Pyqrack

Release v0.18.8 last night passed limited tests of multi-device default layer stack usage before publishing, then stopped working since. I can't explain why, but attempting to diagnose the problem pointed to minor bug fixes and optimizations in this area of the code. Hybrid stabilizer simulation might benefit, at least in the single device `QPager` case. However, multi-device operation under `QUnit` still doesn't typically enable higher qubit widths than single device.

Note that multi-device `QPager` under `QUnit` layer _does_ nominally work in certain easily-attained cases, like 2 "pages" (of maximum allocation segment) apiece on two devices. However, NVIDIA cards typically have 4 page segments that `QUnit` can use already, such that 2-and-2 offers no major improvement over 4 pages on a single device.

This release fixes (at least) the same-platform case of multi-device simulation with default simulation layer stack and options. (There was a minor bug in Qrack that led to `Compose()` under `QUnit` on non-qualified peers, which was hard to test for lack of access to multi-device scales.)

QUnit` buffer commutation cases now include "inversion" buffers with opposite phase factors on their nonzero gate components. (CZ commutation is trickier, but gates that start as inversions already effectively commute, or don't need to, if they can be commuted into phase or inversion gates, because they start as buffered inversions, checked for commutation of inversion on original act of buffering.)

This release also includes minor method parameter typing fixes.

This release adds miscellaneous stabilizer bug fixes, from the current head of Qrack `main` branch.

QUnit` layer over `QStabilizerHybrid` is theoretically capable of transparently achieving stabilizer simulation efficiency on pure Clifford circuits (as well as some near-Clifford circuits). This capability regressed, however. This release fixes it.