Package FastVisionOps for Git installs and organize modules

[Som5ra]

Summary

• build a platform wheel that bundles the compiled native backend
• support direct Git+SSH and Git+HTTPS installation
• organize maintained code under preprocessing, postprocessing, and native packages
• preserve fastvisionops.* and nmss.* compatibility imports
• group standalone historical adapters under legacy/
• verify installed native behavior in CI and document the final layout

Install

python -m pip install "git+ssh://git@github.com/Som5ra/FastVisionOps.git"

Validation

• 47 unit and compatibility tests pass
• clean CPython 3.12 Linux wheel contains the bundled _native library
• fresh Git-URL install from this branch succeeds outside the checkout
• preprocessing: 4.791 ms NumPy vs 0.250 ms native (19.16x)
• 2,500-box NMS: 76.999 ms NumPy vs 17.407 ms native (4.42x)
• 8-image batch: 28.130 ms serial vs 10.990 ms parallel (2.56x)

[Som5ra]

Independent re-review completed with no findings and recommends approval. The initial compatibility regression was fixed in 050d0c3: fastvisionops.native again exports both preprocessing helpers and DEFAULT_OUTPUT, with identity regression coverage. Portable --no-openmp compilation and execution are now verified in CI. Final evidence: 47/47 tests pass, clean wheel and fresh Git-URL installs load the bundled native library, benchmarks validate outputs and report absolute timings, and CI run #10 passes on Python 3.9, 3.12, and 3.13. Approval-equivalent review recorded because GitHub does not allow the PR author to formally approve their own pull request.