fix(compare): retain exported partial and resumed-run provenance

[rudycelekli]

What changed

The JSON exporter preserves partial and resumed-run provenance, but the comparison loader discards those fields. Comparing an aborted report therefore prints provisional scores without the existing warning that they are not comparable to a full run.

Preserve exported partial/abort/not-run and resumed/reused fields, grading metadata and warnings when loading a scorecard. Reuse the existing partial/resumed banners, labeled Baseline or Enhanced. Scores, per-inspection deltas, comparison exit status, cutoff/population policies and resume behavior are unchanged.

Native proof and checks

• Actual public run_selected B01 with the shipped mock provider/customer_support fixture, followed by the actual build_partial_result and generate_json_report producers and public compare CLI. Baseline loses partial=true, the abort reason and missing B06, then displays the provisional60% baseline without its PARTIAL RUN warning. Afterward the inherited warning is present and labeled for the correct report.
• Explicit valid resumed-export metadata is a preservation control: its resume ID/reused count and existing banner survive loading. This does not claim a full resumed-session lifecycle test.
• The ordinary completed selection's public CLI output is byte-identical before and after. No paid/live model call or model-quality claim.
• Before7 regressions fail/2 ordinary controls pass; final19 focused tests pass, including prior merged missing-inspection/unscored comparison controls.
• Signed a02529e: Ruff0.16.9, Bandit, layout and all11 example fixtures pass; Mypy retains1 existing advisory finding.
• Exact official fork Python3.10/3.11/3.12 CI all three jobs pass: https://github.com/rudycelekli/iFixAi/actions/runs/37547032941 . Upstream PR checks are observed separately.

Current main7f1f68b and open source overlaps checked before publication. Merged #132/#126 address missing/unscored comparison rows and null overall scores; this restores exported run provenance. #136 fixes the CI example's separate partial-run exit policy. The owner decision on #143 effective Wilson CI weighting is untouched.

AI assistance: Codex helped investigate, implement and test; the complete source and actual native evidence were independently reviewed before publication. SSH signature and DCO verified.