bump to v1.4.0

BABAPPAlign v1.4.0

This release adds automatic hardware accelerator selection for BABAPPAlign across CUDA, Apple Silicon Metal/MPS, and CPU.

BABAPPAlign now uses --device auto by default and selects the fastest usable backend in this order:

1. CUDA
2. Apple Silicon Metal/MPS
3. CPU fallback

Each accelerator is checked with a small runtime tensor probe before use, so systems where PyTorch reports a backend but cannot actually execute on it will safely fall back to CPU with a warning.

Highlights

• Added Apple Silicon Metal/MPS support
• Added automatic CUDA → MPS → CPU device selection
• Added runtime accelerator probing for safer backend selection
• Updated babappalign and babappascore CLIs to support:
  • --device auto
  • --device cpu
  • --device cuda
  • --device mps
• Removed Conda cpuonly pin to allow accelerator-capable PyTorch builds
• Updated README and installation documentation
• Cleaned release packaging metadata and artifacts
• Added device-selection tests

Validation

Validated locally on Apple Silicon:

• --device auto selected mps
• Protein alignment completed successfully on MPS
• Codon alignment completed successfully on MPS
• babappascore matrix generation completed successfully on MPS
• Test suite passed: 6 passed

The affine-gap dynamic programming alignment core is unchanged. Hardware acceleration affects performance only, not alignment correctness.