fix(analysis): keep matrix rounds distinct in bug-event charts

[grandizzy]

Summary

Matrix/aggregated benchmark runs (the report-matrix path used for foundry PR benchmarks) are flattened by the workflows collector (modules/tests.nu scfuzzbench-collect-all-logs) into a single per-target directory. Each round's log files are renamed to <run_id>-<name>, where <run_id> is the safe-labeled path of that round's logs.zip (so it always ends in logs.zip).

scfuzzbench's analysis only keyed run identity on rel.parts[0] (the directory) and used a single global run_id (infer_run_id finds nothing → "unknown"). As a result every round of a target collapsed into one shared unknown run.

Impact (graphs/tables only — not verdicts)

• bugs_over_time / final_distribution / plateau charts grouped by run_id reported a single run per series → flat/median-of-one charts and "unknown" labels.
• Zero-bug targets (e.g. liquity-v2-governance) were seeded a single unknown:unknown run via events_to_cumulative inventory (which iterated directories, not rounds).
• Verdicts are unaffected: the showmap / relscore / relcov path keys on unique trial files, so rounds stay distinct there and the regression/inconclusive verdicts are computed correctly.

Fix (scfuzzbench only)

• analyze.py: discover_log_files recovers the per-round run_id from the flattened filename prefix; event/throughput/progress parsing use it (falling back to the previous global behaviour for native layouts).
• events_to_cumulative.py: zero-event seeding now walks *.log files (recovering the same per-round prefix) instead of iterating directories, so each round seeds a distinct run and aligns with the event run keys.

Native (non-matrix) and workflows single-run layouts are unchanged (no logs.zip- prefix → same fallback as before).

Tests

• New analysis/tests/test_matrix_run_identity.py pins the contract end-to-end: matrix-flattened layout yields distinct rounds in discover_log_files, parse_logs, and events_to_cumulative (including the zero-bug target), and event/inventory run keys align.
• Full analysis suite passes (3 pre-existing failures in test_runner_command_sanitization are unrelated bash-env issues that also fail on main).

Co-Authored-By: grandizzy 38490174+grandizzy@users.noreply.github.com

Prompted by: grandizzy