feat: write evolution run reports

[steezkelly]

Summary

Implements issue #54 step 2: local, machine-readable promotion artifacts for evolution runs.

Adds:

• evolution/core/run_report.py
• build_run_report(...) for constructing sanitized report payloads
• write_run_report(...) for writing reports/runs/<timestamp>-<target>.json
• artifact diffs via skill.diff
• wiring in evolve_skill.py so successful non-dry-run evolutions write and print a run report path
• failure-path reporting when an evolved skill fails constraints, so failed variants still produce auditable artifacts
• tests that generate reports/diffs without live LLM calls

Report fields include:

• baseline and optimized artifact hashes/sizes
• compatibility aliases: evolved_hash / evolved_size
• dataset source and split counts
• optimizer/eval model names
• constraint results and aggregate pass/fail
• holdout score delta when available
• null cost/latency estimate placeholders when unavailable
• output and diff paths

Safety / privacy notes

The report stores artifact hashes, sizes, metrics, paths, and constraint messages. It does not persist raw session dumps or raw private tool outputs. The diff is the baseline-vs-optimized artifact diff already written for local review.

No remote mutation or PR automation is introduced here.

Test Plan

• RED first: pytest tests/core/test_run_report.py -q failed because evolution.core.run_report did not exist
• pytest tests/core/test_run_report.py -q
• pytest -q
• runtime probe for successful and failed-score report writes
• static added-line security scan
• git diff --check

Result: 142 passed, 11 warnings (DSPy deprecation warnings only).

Partially addresses #54.

[steezkelly]

Closing this split PR in favor of consolidated PR #67. Local integration found review/merge overhead across the stack (notably #61/#64 overlap in evolution/skills/evolve_skill.py), and #67 preserves the combined local test evidence: targeted stack tests 21 passed; full suite 160 passed; GitHub checks were absent on the split PRs. Review #67 instead.