Follow-up: add explicit SAGE coverage for consensus feature union introduced in latest PR

[ypriverol]

Description of the bug

After merging #82 (comet+msgf consensus feature union), we should explicitly validate and (if needed) generalize the same logic for SAGE-involving configurations.

Context

PR #82 fixed the two-engine comet,msgf path by:

• unioning per-engine numeric features,
• applying orientation-aware worst-case imputation,
• adding engine-source indicators.

SAGE support was not explicitly covered in the PR scope/results/tests.

Reference: bigbio/quantms-rescoring#82

Goal

Ensure consensus feature construction behaves correctly for SAGE in:

• search_engines=sage
• search_engines=comet,sage
• search_engines=msgf,sage
• (optional) search_engines=comet,msgf,sage if fully supported

Acceptance criteria

• extra_features contains expected SAGE-derived numeric features in SAGE-involving runs.
• Engine indicator features are present/consistent for all configured engines (including SAGE).
• Missing feature imputation is orientation-aware for SAGE features (or a documented fallback is used).
• No inf/sentinel artifacts leak into model features for SAGE paths.
• Unit tests cover SAGE-only and mixed-engine scenarios.
• Benchmark sanity check at 1% protein FDR shows no regression vs the appropriate single-engine baseline.

Suggested tests

Add coverage in tests/test_consensus_features.py (or equivalent) for:

• single-engine SAGE,
• comet+sage overlap and single-engine subsets,
• msgf+sage overlap and single-engine subsets,
• optional three-engine merge (comet,msgf,sage).

Also validate both:

• ms2features-off (psm_feature_clean) path,
• ms2features-on (annotator) path.

Command used and terminal output

Relevant files

No response

System information

No response