Pytrimal

Changed
- Bumped Cython dependency to `v3.0`.

Fixed
- Overflow in `calculateSpuriousVector` for certain sequence block sizes.
- MMX backend never being used unless explicitly required even when being the best supported backend.
Changed
- Use `archspec` Python package instead of `cpu-features` library for CPU feature detection at runtime.
- Use `importlib.resources.files` to load package data in `pytrimal.tests`.

Fixed
- `calculateSpuriousVector` method of SIMD implementations not being declared `override`.
Changed
- Replaced `aligned_alloc` with `posix_memalign` for compatibility with MacOS.

Added
- AVX2 and MMX implementations of the SIMD statistics computation.
- Tests for all SIMD implementations supported on the local machine.
Changed
- Refactor SIMD code using C++ templates and generic implementation.
Fixed
- Broken rendering of function signatures in Sphinx documentation.
- `residues_mask` and `sequences_mask` attributes of `TrimmedAlignment` not being documented ([1](https://github.com/althonos/pytrimal/issues/1)).

Fixed
- `SimilarityMatrix.nt` inverting the `degenerated` argument value.

Changed
- Replace NEON horizontal sums with implementations using `vaddvq` on Aarch64 and `vpaddl` on Armv7.
- Remove one layer of table lookup in all `Similarity::CalculateVectors` implementations.
- Make all SIMD code use local buffers and deallocate early.

Fixed
- Invalid operator being used in Cython code to deallocate C++ arrays.

Added
- SSE2 and NEON implementations for the `Gaps` statistic.