fix(job-assessment): use unrelated fictional pay numbers and report bad findings cleanly

[darthrootbeer]

Context

Two follow-ups from the WP-9 "still open" list, tracked in JBSR-099. The fictional persona's pay numbers should not match any real person's, and check_findings.py crashed with a Python traceback when a culture list held plain strings instead of objects.

Changes

• Robin Sample's pay figures (floor, min, open ask, target, stretch) and the pay ranges in the three fake postings are now different round fictional values. The order and band relationships are the same, so every score, verdict and trigger is unchanged and fixtures/expected.yaml needed no edit.
• Updated everywhere the old figures appeared: profiles, intake answers, findings, postings, docs examples, inline unit tests, golden files (regenerated, only pay lines differ) and saved run records. Each saved record under tests/ got a one-line note that the figures were changed after the run. No live tests were rerun.
• check_findings.py now reports a list item or section of the wrong shape as an ordinary problem line (for example "culture.strong_positive_phrases[0]: must be an object") and exits 1. A last-resort catch prints one clean line and exits 1 for any shape it cannot name.
• 12 new tests cover the crash shape, other wrong shapes, and the command-line output and exit code.

Review notes

Offline suite: 539 passed, 2 skipped (was 527 passed, 2 skipped; the 12 new tests are the difference).

🤖 Generated with Claude Code

https://claude.ai/code/session_01XV3Ggh86cf2xwUyz28XaN6

• Docs updated