Add comprehensive validation framework for accuracy verification

[skyelaird]

Addresses critical gap: DVOACAP-Python produces output, but we had no way to verify it was CORRECT output. This adds true accuracy validation against reference VOACAP data.

New Files:

1. test_voacap_reference.py

   • Compares DVOACAP-Python predictions against original VOACAP output
   • Uses SampleIO/voacapx.out as reference ground truth
   • Validates SNR, reliability, MUF day factor
   • Reports pass/fail with specific tolerances (±10 dB SNR, ±20% reliability)
   • Usage: python3 test_voacap_reference.py [--hours 1 2] [--freqs 14.15]

2. VALIDATION_STRATEGY.md

   • Comprehensive documentation of validation methodology
   • Explains the difference between "produces output" vs "produces CORRECT output"
   • Documents validation gaps and test coverage status
   • Provides guidance for Claude.ai sessions
   • Outlines future validation approaches (WSPRnet, real-world data)

Changes:

3. README.md - Validation section

   • Clarified distinction between component-level and end-to-end validation
   • Added reference validation instructions
   • Linked to VALIDATION_STRATEGY.md

4. .gitignore

   • Added validation_reference_results.json

Validation Results:

Initial test shows ~55% pass rate with some frequencies returning SNR=0.0, indicating implementation bugs. This is EXACTLY what validation should reveal.

The validation framework now provides:
✅ Deterministic reference comparison (not just "reasonable ranges") ✅ Specific error reporting (which frequencies/hours fail) ✅ Quantitative accuracy metrics
✅ Clear acceptance criteria

This enables proper development workflow:

1. Run test_voacap_reference.py for accuracy
2. Run validate_predictions.py for functional testing
3. Fix bugs revealed by validation
4. Verify improvements don't cause regressions