feat(improvement): user feedback loop for scoring calibration

[bjridicodes]

Let user mark 'good match' / 'bad match' / 'applied' / 'interview'. Feed back into scoring prompts as few-shot examples. Scorer calibrates to user's judgment over time.

[bjridicodes]

Relationships

Depends on:

• feat(tracking): application status states and CLI update tool #34 — feedback requires status states (applied, interviewed, rejected) to know what the user actually did

Enhances:

• feat(quality): semantic deduplication across connectors #43 — user feedback on duplicate or irrelevant matches helps tune deduplication thresholds