Setup precommit hooks and check for coverage levels in CI.

[dlrice]

Purpose

Addresses #162

Tighten the local feedback loop for routine maintenance. Today, lint, type, and test failures only surface after pushing — sometimes only after a PR review reveals a red check. There are no local hooks. Coverage is tracked (since the vitest baseline in #130) but no threshold is enforced, so coverage can drift downward silently between refactors.

This PR adds the mainstream three-tier check pattern (pre-commit / pre-push / CI), shifting routine failures left to local hooks, and wires a coverage-threshold check into CI so coverage can only ratchet upward.

Approach

Three tiers of checks:

1. Pre-commit runs lint-staged against staged files only — ESLint --fix + Prettier --write on .ts/.tsx, Prettier --write on .md/.json/.yml/.yaml. Sub-second on typical diffs.
2. Pre-push runs yarn test — the same lint + types + unit checks CI runs. Mirrors CI exactly so failures show up locally before the push lands.
3. CI adds a Coverage threshold step (yarn test:coverage) using vitest's built-in coverage.thresholds. Initial threshold values are zero placeholders — running yarn test:coverage:ratchet lifts them to the current observed baseline using vitest's --coverage.thresholds.autoUpdate flag. Once seeded, coverage can only go up; CI fails on a drop.

Tooling: Husky 9 + lint-staged (de-facto JavaScript standards). Hooks install automatically via the prepare script — no per-contributor setup. Documented in CONTRIBUTING.md (new "Hooks and feedback loop" section) and README.md (CI section refreshed; the stale hardcoded coverage table replaced with a pointer to vite.config.mjs as the live source of truth).

Bundled minor cleanups: added prettier as a direct devDependency (it was previously a missing peer of eslint-plugin-prettier); added a test:coverage:ratchet script for ratcheting thresholds; updated the README's "Coverage is not collected in CI today" line, which this PR makes obsolete.

Testing

Manual verification (no new unit tests; this is pure tooling):

• yarn install — prepare: husky runs, .husky/_/ infrastructure created.
• Pre-commit smoke test — staged a TS file with a fixable lint issue; git commit invoked yarn lint-staged, the file was auto-fixed, the commit landed cleanly.
• Pre-push smoke test — pushed to a throwaway side branch; yarn test ran end-to-end. Confirmed that introducing a deliberate test failure blocks the push with non-zero exit.
• yarn test:coverage — confirmed vitest produces the coverage report and applies the (currently zero) thresholds without failing.
• yarn test:coverage:ratchet — confirmed vitest rewrites the thresholds: block in vite.config.mjs with the observed values.
• CI workflow — the new "Coverage threshold" step runs in the existing test job after Unit tests. Verified the YAML parses cleanly.

Visual changes

N/A — pure tooling. No UI, render, or build-output changes.

Checklist

• My PR is scoped properly, and "does one thing only" — strictly tooling: hooks, coverage threshold, and the directly-related README/CONTRIBUTING refresh. Other stale items in README (configuration-shape JSON, events block, protvista-uniprot v3/v5 compatibility section) are flagged as a separate v5 docs PR.
• 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 — tests pass. Coverage itself is unchanged (no new tests; this PR adds infrastructure for tracking, not tests). Thresholds are zero placeholders; running yarn test:coverage:ratchet either before merge or in a follow-up commit lifts them to the current baseline.