feat: add job description scoring mode with weighted evaluation and semantic matching

[Kingsam147]

Summary

The existing pipeline evaluates resumes against a fixed rubric hardcoded for a single role (Software Intern at HackerRank). This PR adds a second evaluation mode that accepts any job description and scores the resume against it using a 7-category weighted model, making the tool useful for any role.

At startup the user is prompted to choose between the two modes. The original HackerRank scoring is untouched.

How it works

Input: paste any job description into job_description.txt in the project root, place resume.pdf in the same directory, then run python score.py and select mode 2.

Both files ship as empty placeholders in the repo so new users know exactly where to put their inputs. To prevent personal content from ever being staged or pushed, run once after cloning:

git update-index --skip-worktree job_description.txt resume.pdf

Pipeline (mode 2):

1. LLM extracts structured requirements from the job description (job title, required skills, preferred skills, years of experience, education requirements, must-have qualifications)
2. Sentence Transformers (all-MiniLM-L6-v2) computes cosine similarity between the job description and resume embeddings for semantic/keyword matching
3. LLM scores six categories against the extracted requirements and full job description text
4. Weighted total is computed in Python

Scoring weights:

Category	Weight
Skills Match	30%
Experience Match	20%
Keyword & Semantic Match (Sentence Transformers)	15%
Job Title Alignment	10%
Education & Certifications	10%
Resume Quality	10%
Missing Critical Requirements	5%

Results are written to job_evaluations.csv in development mode (separate from the existing resume_evaluations.csv).

Changes

New files:

• job_description.txt — empty placeholder for the job description input
• resume.pdf — empty placeholder for the candidate resume
• prompts/templates/job_description_extraction.jinja
• prompts/templates/job_evaluation_criteria.jinja
• prompts/templates/job_evaluation_system_message.jinja

Modified files:

• models.py — JobDescriptionData, JobCategoryScore, JobScores, LLMJobEvaluationResponse, JobEvaluationData Pydantic models
• evaluator.py — JobDescriptionEvaluator class
• score.py — mode selector, fixed resume.pdf path, routing, new output formatter, job_evaluations.csv writing; resume cache now invalidated automatically when resume.pdf is replaced (mtime comparison)
• transform.py — transform_job_evaluation_response(), removed stale pdb import
• prompts/template_manager.py — registers three new templates
• requirements.txt — adds sentence-transformers
• .gitignore — adds job_evaluations.csv, package-lock.json; removes resume.pdf now that it is a tracked placeholder

Test plan

• Validated with a real resume in mode 1 (original HackerRank scoring) — output unchanged from pre-PR behaviour
• Validated with a real resume + job description in mode 2 using Ollama (gemma3:4b)
• Validated with a real resume + job description in mode 2 using Gemini (gemini-2.5-flash)
• Empty job_description.txt in mode 2 exits with a clear error message
• Missing resume.pdf exits with a clear error message
• job_evaluations.csv is created and appended to correctly in development mode
• Personal resume and job description content cannot be accidentally staged or pushed (--skip-worktree)
• Replacing resume.pdf with a different file triggers automatic cache invalidation on the next run