Phase 1: V3 enhancements - parameter errors, fit_intercept, stability fixes

New Features:
- Added analytical parameter error estimates (SE for all coefficients)
- Added fit_intercept parameter (True/False for intercept vs through-origin)
- Enhanced Inf/NaN filtering before matrix operations
- Enhanced diagnostics (4 status levels: OK, ILL_CONDITIONED_RIDGED, etc.)

Critical Bug Fixes:
- Fixed RMS calculation to use degrees of freedom (unbiased variance)
- Fixed weighted variance (was using w³ instead of w)
- Fixed diag=False to consistently suppress all diagnostic columns

Testing:
- Added 8 comprehensive test functions for V3 features
- All 32 tests passing (19.93s)
- Validated by GPT and Gemini reviews

Backward Compatibility:
- fit_intercept=True (default) preserves existing behavior
- All existing tests pass unchanged
- Float64 enforcement for numerical stability

Author: miranov25