Pylops

* Modified all operators and solvers to work with cupy arrays
* Added ``eigs`` and ``solver`` submodules to ``optimization``
* Added ``deps`` and ``backend`` submodules to ``utils``
* Fixed bug in ``pylops.signalprocessing.Convolve2D`` and
``pylops.signalprocessing.ConvolveND`` when dealing with
filters that have less dimensions than the input vector.

* Fixed import of ``pyfttw`` when not available in
``pylops.signalprocessing.ChirpRadon3D``

* Added ``pylops.signalprocessing.ChirpRadon2D`` and
``pylops.signalprocessing.ChirpRadon3D`` operators.
* Fixed bug in the inferred dimensions for regularization data creation
in ``pylops.optimization.leastsquares.NormalEquationsInversion``,
``pylops.optimization.leastsquares.RegularizedInversion``, and
``pylops.optimization.sparsity.SplitBregman``.
* Changed dtype of ``pylops.HStack`` to allow automatic inference from
dtypes of input operator.
* Modified dtype of ``pylops.waveeqprocessing.Marchenko`` operator to
ensure that outputs of forward and adjoint are real arrays.
* Reverted to previous complex-friendly implementation of
``pylops.optimization.sparsity._softthreshold`` to avoid division by 0.

* Added ``tosparse`` method to ``pylops.LinearOperator``.
* Added ``kind=linear`` in ``pylops.signalprocessing.Seislet`` operator.
* Added ``kind`` to ``pylops.basicoperators.FirstDerivative``.
operator to perform forward and backward (as well as centered)
derivatives.
* Added ``kind`` to ``pylops.optimization.sparsity.IRLS``
solver to choose between data or model sparsity.
* Added possibility to use ``scipy.sparse.linalg.lobpcg`` in
``pylops.LinearOperator.eigs`` and ``pylops.LinearOperator.cond``.
* Added possibility to use ``scipy.signal.oaconvolve`` in
``pylops.signalprocessing.Convolve1D``.
* Added ``NRegs`` to ``pylops.optimization.leastsquares.NormalEquationsInversion``
to allow providing regularization terms directly in the form of ``H^T H``.

* Changed internal behaviour of ``pylops.sparsity.OMP`` when
`niter_inner=0`. Automatically reverts to Matching Pursuit algorithm.
* Changed handling of `dtype` in ``pylops.signalprocessing.FFT`` and
``pylops.signalprocessing.FFT2D`` to ensure that the type of the input
vector is retained when applying forward and adjoint.
* Added `dtype` parameter to the `FFT` calls in the definition of the
``pylops.waveeqprocessing.MDD`` operation. This ensure that the type
of the real part of `G` input is enforced to the output vectors of the
forward and adjoint operations.

* Added ``pylops.waveeqprocessing.Deghosting`` and
``pylops.signalprocessing.Seislet`` operators
* Added hard and half thresholds in ``pylops.optimization.sparsity.ISTA``
and ``pylops.optimization.sparsity.FISTA`` solvers
* Added ``prescaled`` input parameter to ``pylops.waveeqprocessing.MDC``
and ``pylops.waveeqprocessing.Marchenko``
* Added sinc interpolation to ``pylops.signalprocessing.Interp``
(``kind == 'sinc'``)
* Modified ``pylops.waveeqprocessing.marchenko.directwave`` to
to model analytical responses from both sources of volume injection
(``derivative=False``) and source of volume injection rate
(``derivative=True``)
* Added ``pylops.LinearOperator.asoperator`` method to
``pylops.LinearOperator``
* Added ``pylops.utils.signalprocessing.slope_estimate`` function
* Fix bug in ``pylops.signalprocessing.Radon2D`` and
``pylops.signalprocessing.Radon3D`` when ``onthefly=True`` returning the
same result as when ``onthefly=False``