feat(inference): unify native and exported model prediction

[leeclemnet]

Summary

Exported RF-DETR models require runtime-specific prediction code today. RFDETRInference(source).predict() accepts native models, checkpoints, and all eight export formats and returns the same Supervision results as native prediction.

Changes

• Add an inference-only facade with lazy, format-owned loaders that reuse existing runtime execution.
• Share image preparation and result decoding between native and exported prediction.
• Save versioned inference metadata with exports and validate it when artifacts load.
• Add public export → load → predict coverage to the existing format parity suites.
• Document the common prediction API and explain the role of existing raw runtime tools.

Verification

• Add public export → load → predict coverage to the existing format parity suites.

Performance

n/a

Related issues

n/a

[codecov]

Codecov Report

❌ Patch coverage is 88.70813% with 118 lines in your changes missing coverage. Please review.
✅ Project coverage is 90%. Comparing base (8913f7b) to head (bc62191).

Additional details and impacted files

@@           Coverage Diff            @@
##           develop   #1563    +/-   ##
========================================
- Coverage       90%     90%    -0%     
========================================
  Files          138     145     +7     
  Lines        18264   19063   +799     
========================================
+ Hits         16436   17128   +692     
- Misses        1828    1935   +107     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Borda]

this needs better API design and resonsildation of already prenet inference classes 🦝