v0.10.0

Version 0.10.0

Highlights:
- BREAKING: Unified Tensor API — TensorImage removed, Tensor is the
  single type across Rust, C, and Python
- dtype parameter on create_image() for direct int8 tensor output
- Int8 GPU shaders eliminate CPU XOR post-pass (~2× faster int8 pipeline)
- Dual Python wheels (abi3-py311 zero-copy + abi3-py38 fallback)
- TensorDyn API expansion (new, from_fd, size, memory, reshape, clone_fd)
- Multiple int8 letterbox and alpha-channel bias fixes

See CHANGELOG.md for full details.