Better entropy mix-in for OTOC (2)

From further empirical inspection, it eventually became obvious that, for every single additional Z or Y operator added to the OTOC Pauli strings, the "signal fraction" of the heuristic could be about halved, (in other words, mixed in with half maximal superposition). Further, conceptually, X basis measurement should produce a maximal superposition on perfect reversal of TFIM without perturbation, not deterministic |+> state. After these improvements, results are nearly optimal, until entropy fraction completely dominates and XEB and HOG probability would be measuring variation that's basically in the margin of error, for deviation from a flat distribution. (The point of RCS is to recreate exactly that marginal variation; "utility" or "advantage claims" should probably rest on distributions that can't be meaningfully understood for practical purposes to be just flat, in the mean field.)

Full Changelog: v9.5.6...v9.5.7

sha1sum results:
9a2f28311a89ac50da9955b3bd7729122198ee47 pyqrackising-9.5.7-py3-none-macosx_14_0_arm64.whl
be60074c91dffe03609ab8b295000439ce29daa9 pyqrackising-9.5.7-py3-none-macosx_15_0_arm64.whl
654db805884cd693989a0d78257238d17098ed2c pyqrackising-9.5.7-py3-none-manylinux_2_35_x86_64.whl
d0de0b0967ba4dfd420c0be355024e85ad7edd1f pyqrackising-9.5.7-py3-none-manylinux_2_39_x86_64.whl
2f6da83f630e69619f2117c7731236201853d8fd pyqrackising-9.5.7-py3-none-win_amd64.whl
03507e6edd54e370530c446133bc084bb269fc64 pyqrackising-9.5.7.tar.gz