[codex] Add batch lint memory benchmark

[luojiyin1987]

This was generated by AI during triage.

Summary

• add a Linux peak-memory benchmark for batch lint
• generate temporary Markdown fixtures with configurable size and worker count
• support repeated runs with machine-readable RSS and elapsed-time summaries
• add a small smoke test and npm commands
• reject unsupported platforms with a clear error

Why

Batch lint currently loads all Markdown content before linting. This PR establishes a repeatable baseline before the production data path is changed in a later PR.

Baseline

Node.js 24.15.0, 16 files, 1,000,010 bytes per file, three runs per worker setting:

Threads	Median peak RSS	RSS range	Median wall time	Time range
1	972,620 KiB	971,028–1,051,080 KiB	11.17s	9.87–11.45s
4	2,452,408 KiB	2,198,368–2,570,680 KiB	4.38s	4.08–4.48s

Platform scope

This benchmark intentionally supports Linux with GNU /usr/bin/time -v. macOS uses BSD time with different flags, output fields, and units. macOS support is left to a volunteer contribution that can include BSD output fixtures, unit-conversion tests, and real-device verification.

Impact

No production source or CLI behavior changes. The benchmark removes its temporary fixtures in a finally block.

Validation

• npm test -- --runInBand: 45 tests passed
• npm run build: passed
• npx eslint --ext .ts,.tsx ./: passed
• npm run test:benchmark-memory: passed
• node --check scripts/benchmark-memory.mjs: passed
• invalid --runs 0: exited non-zero
• git diff --check: passed

Closes #69