Performance baseline

[dlrice]

Addresses:
Performance: define baseline metrics and profiling workflow #100

Adds bench/ for repeatable performance baselines. Three layers, no src/ changes:

• Bundle size — raw + gzipped totals from dist/.
• Lighthouse CI — vite preview against three UniProt scenarios (P05067, P38398, A0A2K5ULD0), 5 runs/URL.
• Custom milestones — bench/instrument.js (loaded on ?bench=1) emits fetch-and-parse / render / total via MutationObserver; Lighthouse captures them via user-timings.

Adds @lhci/cli devDep.

Purpose

Repeatable, documented perf measurement so the upcoming config refactor can be evaluated against a stored baseline. Satisfies the perf-baseline issue's acceptance criteria.

Approach

yarn bench runs all three layers and writes bench/results/summary.md. Reference snapshots are committed under bench/baselines/; per-run output is gitignored.

Custom marks observe DOM transitions on the host (loader removed → data loaded; nightingale-manager present → first render; 250 ms of subtree quiescence → tracks settled).

Testing

• Ran yarn bench end-to-end; values stable, scenarios differentiate as expected (fetch-and-parse scales with payload; TBT catches canvas-paint cost the observer can't see).
• node --check on all scripts; smoke-tested summarize.mjs against a synthetic LH report.
• Captured bench/baselines/summary-pre-config-<sha>.md.

Visual changes

None. Demo renders identically without ?bench=1.

Checklist

• My PR is scoped properly, and "does one thing only"
• I have reviewed my own code
• I have checked that linting checks pass and type safety is respected
• I have checked that tests pass and coverage has at least improved, and if not explained the reasons why
• If needed, the changes have been previewed by all interested parties.