Improve robustness, scoring, and DX: CLI flags, Live Demo recognition, rate-limit handling, cache validation, docs/ignoresfeat(cli): add --force/--no-github/--max-workers; feat(projects): cap…

[abhiram050904]

Summary:
This PR strengthens the resume scoring pipeline, adds reliability under rate limits, improves project attribution (Live Demos), and cleans up developer ergonomics and docs.
Features:

• CLI: added --force, --no-github, --max-workers to control caching, enrichment, and concurrency.
• Projects: capture both repo_url and live_url; evaluator now sees “Repo URL” and “Live Demo”.
• Languages: new extraction path and model wiring for spoken languages.
• Reliability:
• Rate-limit resilience: parsed backoff on 429s, configurable concurrency, targeted retries, and sequential fallback when parallel fails.
• Cache validation: resume cache now stores file hash, model, and template hashes; auto re-extracts if stale/empty; --force bypasses caches.
• Heuristic fallbacks: fills skills, projects, and languages when LLM extraction is sparse.
• GitHub data: robust cache handling (invalid/empty cache refresh), token guidance, and rate-limit-aware behavior.

Fixes:

• Live demo deductions addressed by explicitly surfacing Live Demo links.
• Pydantic safety: avoid instance model_fields access when checking emptiness.
• Models: added LanguagesSection and extended Project with repo_url/live_url; consistency improvements.
• Docs/Config:
• README: documented flags and usage; clarified caching behavior.
• .env.example: sanitized placeholders only; added GITHUB_TOKEN.
• .gitignore: ignore .pdf, cache/, .venv_win/, python-.exe, etc.