v0.5.0

Added

• resume-batch: tailor across a whole jobs/ folder in one pass. Triages postings for
  genuine fit (keyword searches drag in unrelated roles — those get filtered, not
  tailored), clusters the survivors by the shape of resume they need rather than by job
  title, so near-identical roles share one variant instead of spawning a throwaway each.
  The user picks scope (all clusters vs. top-N by fit); each selected cluster yields a
  resume-<cluster>.tex, each job a company-specific cover letter, and jobs/index.md
  records the whole search. Never edits the master.
• resume-scrape: collect job postings from a listing page (e.g. a LinkedIn jobs
  search) into a local jobs/ folder for batch tailoring. Uses a human-in-the-loop
  login — it opens a visible browser, the user logs in themselves, then Claude drives
  the already-authenticated session to gather each posting's title, company, location,
  and description. Scraped content is treated as untrusted data, never instructions.
• Vendored the gstack browse subsystem (MIT) under vendor/browse/ as the browser
  engine for resume-scrape. Source-only: deps and Chromium install on first run via
  bun, mirroring how the LaTeX server builds on first use with uv. Adds bun and
  node as prerequisites for the scraping workflow only.
• skills/resume-scrape/scripts/linkedin-extract.js: the posting extractor, anchored on
  document.title, the "About the job" heading, and data-testid hooks rather than
  LinkedIn's hashed, server-driven-UI class names. Verified against a live logged-in
  search: 25/25 postings captured with full descriptions.