Fix init_thresholds()

Hamming weight initialization to equal superposition, at the start of adiabatic TFIM, was off 64-bit precision range, hence it shouldn't use numba JIT. The most precise way to handle it is with math.comb() and ratios of arbitrary precision integers made floating-point, hence we should rely on default Python language math instead. (Performance remains acceptable.)

Full Changelog: v9.5.3...v9.5.4

sha1sum results:
1ed9d301268abf93e8c8baf391ff8e22c5d26e04 pyqrackising-9.5.4-py3-none-macosx_14_0_arm64.whl
c3f0c7d58a661801c65007f0bec92cd86a72fcd0 pyqrackising-9.5.4-py3-none-macosx_15_0_arm64.whl
58984710d0995b89afa024ca098abf32edada461 pyqrackising-9.5.4-py3-none-manylinux_2_35_x86_64.whl
0502b81e3259b458aa10c8f28edb25448da8f6f0 pyqrackising-9.5.4-py3-none-manylinux_2_39_x86_64.whl
55d004a65dcc5286a1a8b8d9391dd58debd636f2 pyqrackising-9.5.4-py3-none-win_amd64.whl
f617e77c10f6cf8db6852c11a85a96ce6f13b5c2 pyqrackising-9.5.4.tar.gz