AI resume parsing & JD matching

1. Open a job in /admin/recruitment.
2. Click Parse all resumes — every unparsed application is scored against the JD.
3. Each application now shows a 0-100 match score + skill match breakdown.
4. Sort the application list by score to surface the top 10% in seconds.
5. If you update the JD (requirements / experience range / skills), click Rescore to refresh scores without re-parsing resumes.

• Contact: name, email, phone
• Skills: tech + functional (~60 keywords + AI-extracted in upgrade mode)
• Experience: years of total experience
• Education: degree + institute + year (best-effort)
• Current role and a short summary

0-100 with three components:

• Skills (up to 70 pts): % of job's required skills that appear in the resume.
• Experience (up to 25 pts): full marks when candidate's years fall in the job's range; penalised below or way-above.
• Presence (up to 5 pts): bonus for valid email, phone, education entry.

Skill source: comma-separated Job.skills field first; if empty, any heuristic-detected skill that also appears in requirements / description.

Heuristic mode (always on): regex + curated 60-word skill bag. Free, instant, ~70% accuracy on text resumes.

Gemini upgrade (when GEMINI_API_KEY is set): structured JSON extraction via Gemini 2.0 Flash. Better at synonyms, niche skills, and free-form education sections. Falls back silently if the API call fails.

• PDF text extraction is best-effort — text-PDFs work, scanned/image-only PDFs don't until an OCR pass is added.
• For non-text resumes, paste the text via the "Parse with pasted text" option.
• Match scoring is deterministic on the inputs — same JD + same resume = same score every run. Useful for audit / bias review.