Pyqrack

With this release, if Qrack OpenCL kernels are precompiled, then loading of source will not be attempted unless binary loading fails. Previously, OpenCL kernel source was loaded, regardless. (Further, Qrack switched standard convention from preferring post-fix increment to preferring prefix, when the return value is unused, but this likely makes no effective difference in the compiled binary.)

Additional refactoring and suggestions have been incorporated from a community contributor, with thanks!

A community member has been kind enough to provide bug fixes and standards improvements, under Daniel Strano's review. (See https://github.com/vm6502q/pyqrack/pull/18, although `staticmethod` definitions turn out not to be C-callable, and this has already been tested and corrected, with thanks!)

Since that pull request motivated a release, I also iterated Qrack binaries, for the change differential since last release. Those changes comprise Qrack code standards changes, regarding explicit integer type literals. Likely, these changes make no logical difference to outputs, but the point was to proactively guard against overflow conditions, at or around primitive type maximum values.

A (long-standing) bug in the Schmidt decomposition layer implementation of the `fsim()` gate has been fixed. Further, an edge case in `muln()`/`divn()`/`cmuln()`/`cdivn()` has been fixed, which could have become pertinent when output input and output register sizes were mismatched to the next highest power of two of the modulus.

Modular multiplication and division (both with and without control qubits) are no longer limited to powers of two, for modulus. These are gate-level decomposition implementations, rather than custom OpenCL kernels, so they can work with the full breadth of simulation methods, including `isPaged` and `isSchmidtDecompose`.

This release adds the `force_m()` method to `QrackSimulator`, for a (non-physical) option to force the outcome of probabilistic methods.

This fixes a pointer edge case and a variable name typo bug in the PInvoke API, only. Apologies for my scattered manner; I'm one developer. Build width is 2^12 across all binaries.