Optimize differential coverage analysis

[decofe]

Summary

• stream AFL showmap parsing instead of loading each file with read_text().splitlines()
• compute relscores from precomputed per-approach edge hit counts instead of scanning every trial for every edge repeatedly
• compute relcov from precomputed reference unions instead of repeatedly constructing DifferentialCoverage/ApproachData views
• remove the now-unused differential-coverage analysis dependency

Prompted by: @0xalpharush

Validation

• git diff --check
• uv run python -m py_compile analysis/analyze.py scripts/run_analysis_filtered.py
• uv run --with-requirements analysis/requirements.txt python -m unittest analysis.tests.test_analyze_log_filtering analysis.tests.test_analyze_progress_metrics_parser analysis.tests.test_analyze_throughput_parser analysis.tests.test_differential_coverage analysis.tests.test_foundry_run_showmap_args
• uv run --with-requirements analysis/requirements.txt --with differential-coverage python - <<'PY' ... synthetic parity/perf check: old dependency path 0.135997s median vs optimized local path 0.003102s median on a 3-approach / 25-trial / 80-edge campaign