Pyqrack

The `QUnit` class in the underlying Qrack library has been refactored. The upshot for PyQrack users is that package binaries are smaller, and Schmidt decomposition techniques extend beyond 2 controls to arbitrary control width gates. (Schmidt decomposition in general should be more reliably and systematically applied.)

A particular optimization case of `CNOT` in the underlying Qrack library that was bugged on edge cases has been cut, fixing the edge case.

Also, a new stabilizer optimization case has been added for controlled phase gates.

Users might want to instantiate a simulator that does _only_ traditional "ket" or "state vector" simulation. I have fixed a bug that previously caused a segmentation fault when attempting to instantiate such a simulator.

Windows build of Qrack were previously bugged on hybrid stabilizer simulation, which has been completely fixed. (The reason was use of C++ `std::swap()` with `std::vector<bool>`, which is strictly a violation of the language standard, though _only_ the Windows compiler is strictly adherent to the explicit standard, as far as I can tell.)

A bug in the `QrackSimulator` constructor has also been fixed; this release now allows non-default constructor Boolean options.

The Qrack binary has changed its convention slightly on the `QRACK_GPU_OFFSET_QB` environment variable: strictly greater-than values for processing element count round _up_ to the next power of 2, now, instead of rounding _down_, and the default value of the environment variable has been correspondingly reduced by 1.

Performance improvements in this release include OpenCL work item resizing, achieving ~2x on quantum Fourier transform performance on a test system, and (internal) stabilizer-to-ket conversion speed improvement.

In addition to several minor performance improvements, PyQrack v0.10.0 adds the `isHostPointer` constructor kwarg to `QrackSimulator`. This flag, when true, will attempt to force allocation of OpenCL objects in general host RAM, instead of potentially VRAM. In cases, this might help support wider qubit count simulations.