feat(job-assessment): wire end to end and retire job-fit-screen

[darthrootbeer]

Context

The job assessment pieces were merged one by one but never run together, and the old early scorer in tools/ still sat beside them with a banner saying it would be replaced. This PR runs the whole chain end to end and removes the old tool. Ticket: JBSR-099.

Changes

• New scripts/assess_offline.py runs check, score, verdict, render, write, rename and email card with no model and no network.
• New tests/test_e2e_offline.py runs it on the three fictional postings and compares to expected.yaml and fixtures/golden/ (28 tests; full suite 461 passed).
• New tests/live/ (skipped unless JA_LIVE=1 and the claude CLI exists) with saved records: tests/intake-transcript.md, tests/e2e-output.md, and the first attempt tests/e2e-output-run1.md, which got one verdict wrong.
• CI gets a clean-clone job: fresh clone, venv, validator, offline chain on all three postings, full suite, clean tree.
• Robin's scripted intake answers gain the name, target roles and years answers.
• tools/job-fit-screen/ is removed (git history keeps it). Top-level README keeps docs-pipeline as item 1 and adds the job assessment system as item 2; ROADMAP and the job-assessment status list are updated.

Checks

• README.md and ROADMAP.md describe what is in the tree after this change
• safety_scan range scan is clean
• Tests were run, real output is in the log
• Private gate: cet-gate.sh is not available in this session, not run

🤖 Generated with Claude Code

https://claude.ai/code/session_01XV3Ggh86cf2xwUyz28XaN6

• Docs updated