perf: first ubuntu-latest baseline

[saldestechnology]

Summary

Commits the first perf/baselines/ubuntu-latest.json, enabling the CI regression gate (median > baseline * 1.20) that has so far been warn-passing for lack of a baseline. Also updates perf/baselines/README.md to record capture provenance in place of the "no baseline committed yet" note.

Capture provenance

• CI run: 29126075011 (ubuntu-latest, 2026-07-10)
• Commit: 98e06386afcf0ec02df8ba13208bdf9444050bf3
• Suite: full, budget scale 1.5, perf profile, fixture_format_version 1

Scenario	Median	Max RSS
check_against_head	162.8 ms	55,812 kB
index_cold_150k	13,641.7 ms	93,652 kB
index_cold_2k	8,043.8 ms	67,500 kB
index_incremental_1	209.7 ms	47,616 kB
map_cached	46.2 ms	42,552 kB
score_3_changed	5,965.1 ms	280,368 kB
sql_v1_query	266.2 ms	93,080 kB

All seven scenarios are included, deliberately:

• index_cold_2k has no absolute budget (informational) but still benefits from regression tracking.
• score_3_changed currently exceeds its absolute budget (2000 ms, 3000 ms at scale 1.5) — tracked in perf: ctx score --against HEAD is 3x over its hook-path budget on the 2k-file fixture #23. The baseline is the regression reference and is independent of the budget gate; the perf job stays advisory until perf: ctx score --against HEAD is 3x over its hook-path budget on the 2k-file fixture #23 is fixed.

Verification

• Baseline file matches the serde model in perf/src/compare.rs (Baseline { meta { profile, fixture_format_version, budget_scale_at_capture, commit }, scenarios }); shape asserted with a one-off script.
• cargo test --manifest-path perf/Cargo.toml — 20 tests pass.

🤖 Generated with Claude Code