Aerosandbox

- In general, uses NeuralFoil throughout instead of XFoil. `asb.Airfoil` no longer needs to have its `generate_polars()` method called before assembling an airplane object. In general, the old "polar functions" of `Airfoil` (e.g., `Airfoil.CL_function()`) are no longer used.
- Added `asb.Airfoil.get_aero_from_neuralfoil()`, which is the standard rapid airfoil aerodynamics analysis method in AeroSandbox now.
- Significant backend change in `asb.AeroBuildup` to use NeuralFoil instead of XFoil for airfoil analysis. This is a significant mathematical change, but it should be transparent to the user. Nevertheless, bumping minor version because this change has the potential to cause differences in power-users' workflows (i.e., if they're hacking Airfoil.xfoil_data or Airfoil.generate_polars() to make custom polars).
- Made a change in `asb.AeroBuildup` within the `.fuselage_aerodynamics()` method that causes fuselage potential flow lift (via slender body theory) to act in the local-wind-normal direction, not halfway between the body-axes-$z$ and wind-axes-$z$ vector as before. The new implementation agrees with Drela's Flight Vehicle Aerodynamics (and intuition), but disagrees with Jorgensen 1977, a NASA TR on fuselage aero. In general, new aero predictions are verified as more realistic (no longer generate a suction force from potential flow in edge cases, satisfying d'Alembert). In general, this results in slightly higher (~5%) fuselage drag prediction, and negligible changes to fuselage lift prediction. For most airplanes this will make negligible difference, but for lifting-body-type aircraft this may be perceptible. As a bonus, optimization should be better-behaved through `asb.AeroBuildup` analyses with fuselages now.
- Added lots of new tools for computing drag of various miscellaneous components in `aerosandbox.library.aerodynamics.components`.

-----

- Fixed a bug in `asb.Airfoil.add_control_surface()` that resulted in duplicated leading-edge nodes.
- Added new ability to draw shaded 3D renderings of airplanes with Matplotlib backend, using Matplotlib 3.7.0 features. Available through `asb.Airplane.draw_three_view()` or `asb.Airplane.draw(backend="matplotlib", ...)`.
- Recast the Kulfan airfoil parameterization problem as a least-squares problem, resulting in a 20x speedup in converting an airfoil coordinate set to CST parameters. Available in `aerosandbox.geometry.airfoil.airfoil_families` in `get_kulfan_parameters()`. New behavior is added as the default, but the old behavior is still available as `get_kulfan_parameters(method="opti")` (as opposed to `"least_squares"`).

- Added lots of new tools and refinements to support airfoil conversion between raw coordinates and Kulfan (CST) coefficients, in `aerosandbox/geometry/airfoil/airfoil_families.py`. Conversions are now bidirectional to/from coordinates and CST coefficients.
- Lots of modifications to the airfoil database stored in `aerosandbox/geometry/airfoil/airfoil_database`. A few (<5) particularly unusable airfoils were removed. With many others (~50), airfoil coordinates were tweaked to prevent self-intersection (i.e., parts of the airfoil where the upper and lower surfaces cross each other). Generally, modifications were near the trailing edge, and on the lower surface when possible (less sensitive boundary layer), and no more than 0.1% of the chord. Several airfoils with duplicated coordinates had these removed. Several airfoils that were egregiously not normalized (e.g., scaled for x/c from 0 to 100, not 0 to 1) were rescaled appropriately. Added new unit tests for airfoil database validity.
- Reworked the `asb.Airfoil.repanel()` method to use `scipy.interpolate.CubicSpline` rather than `PChipInterpolator`. This allows for more precise control over LE/TE boundary conditions, so that it is less likely (but not impossible) to produce self-intersecting geometries when repaneling very-low-resolution airfoils. Upsampled airfoils will be ever-so-slightly different, which may have miniscule but nonzero changes on existing code. Also rewrote code for better readability.
- Improved the plate buckling structural model in `aerosandbox/structures/buckling.py` to take into account Poisson ratio.

- Added turboshaft modeling capabilities in `aerosandbox.library.power_turboshaft`. Added turbine engine database in `aerosandbox/library/datasets/turbine_engines`.
- Added tools for uncertainty quantification of time-series datasets.
- BREAKING: Renamed `aerosandbox.library.propulsion_jet` to `propulsion_turbofan` for clarification. Very minor change, as the old module only contained one function (`mass_turbofan`) that was not used in any other modules or known production code.
- Added turbofan/turbojet modeling capabilities in `aerosandbox.library.power_turbofan`.
- Better docs throughout.

- Added first experimental ability to support black-box functions in optimization, available in `aerosandbox.modeling.black_box`. This is a very early prototype, and the API is subject to change.
- Fixed a bug in `asb.Airplane` auto-generation of s_ref, c_ref, and b_ref that would cause hard-coded s_ref values in the constructor to not be applied to the Airplane instance. Also, adds the capability for auto-generation of reference values from fuselages, if no wing is present.
- Fixed a bug within `asb.AeroBuildup` effective span calculation for symmetric wings that were used to represent doubled vertical stabilizers. Now, if you create an aircraft with twin vertical stabilizers, the Trefftz-plane wake will correctly not carry over between the two vstabs. (If you have a symmetric horizontal stabilizer, it will still carry over - currently this is a function of effective dihedral angle.)
- Fixed Torenbeek wing weight model to use a simpler model for speedbrake / spoiler weight. This resolves a (possible?) error in the wing weight model due to an ambiguity in the source text (Torenbeek's "Synthesis of Subsonic Airplane Design", Appendix C) about what area they were referring to.
- Fixed a bug with uncertainty bootstrapping in `aerosandbox.tools.pretty_plots`, when normalizing input data.

- BREAKING: Critical bugfix in `asb.OperatingPoint.compute_rotation_matrix_wind_to_geometry()`, which in turn affects `asb.OperatingPoint.compute_freestream_direction_geometry_axes()`: A sign was reversed in the rotation matrix calculation for the sideslip angle beta. This resulted in the sign convention for beta (sideslip) being flipped, ultimately causing a sign error. In practice, this bug affected `asb.VortexLatticeMethod` analyses with nonzero beta, causing flipped signs for `CY, Cn`, etc. This bug was introduced in AeroSandbox v4.0.0, and was present in all versions of AeroSandbox v4.0.0-v4.0.6. This bug did not affect `asb.AeroBuildup` or any other aerodynamics analysis methods. (Thanks to carlitador for catching this.)
- Added `asb.AVL.open_interactive()`, to interactively launch an AVL session.
- Improved `__repr__` methods throughout for better readability.
- Updated `asb.AeroBuildup` to add induced drag on a whole-airplane level, not per-lifting-object. In general, this will result in slightly higher induced drag, and also improves optimization pressure - for example, tandem-wing configurations are no longer unrealistically attractive, since the induced drag scales superlinearly with respect to total lift.
- Improved `asb.Airfoil.generate_polars()` to not error out when a cache filename is provided where the containing directory does not yet exist. (Instead, it now creates the directory.) Fixed a bug in `generate_polars()` that now allows any Reynolds number input list to be specified when calling XFoil.
- Added `asb.Airfoil.plot_polars()` to make polar functions more interpretable.
- Added an optional `color` attribute to `Wing`, `Fuselage`, and `Propulsor` objects, which can be set during instantiation or manually set afterwards. This will control the color that the component is drawn with during visualization; currently only works with the `Airplane.draw_wireframe()` and `Airplane.draw_three_view()` methods, but will be extended to others later.
- Updated pinned CasADi version to 3.6.1. Note that CasADi MUST be v3.6.0 or higher, otherwise automatic differentiation errors may occur due to undefined primitives. This shouldn't be an issue, as this is now pinned in ASB's setup.py.