fix: classify phantom references under AI quality

[7arnav1]

Summary

Adds a dedicated AI_QUALITY negative-context category and uses it for phantom-reference findings, as proposed in #473.

Problem

Phantom references identify hallucinated symbol usage from AI-assisted coding, but they were exported under the generic "Completeness Anti-Patterns" section. This diverged from signal_registry.py, which already categorizes PHR as ai_quality, and made it impossible for agent consumers to filter AI-specific constraints from negative-context output.

Changes

• Added AI_QUALITY = "ai_quality" to NegativeContextCategory (models/_enums.py)
• Mapped SignalType.PHANTOM_REFERENCE to AI_QUALITY in _SIGNAL_CATEGORY (negative_context/core.py)
• Added the AI-Generated Code Anti-Patterns heading to _CATEGORY_HEADING (negative_context/export.py)
• Updated _gen_phr in generators.py, which hardcoded COMPLETENESS
• Updated the existing PHR category assertion in tests/test_negative_context.py
• Added tests/test_issue_473_ai_quality_category.py: the mapping, the heading, a completeness check that every category has a heading, and end-to-end tests that a PHR finding produces an AI_QUALITY item rendered under the new section
• Added ai_quality to the category list in docs-site/reference/negative-context.md

Checked drift.output.schema.json / drift.schema.json — neither enumerates NegativeContextCategory values, so no schema regeneration is needed.

Note for consumers

Raw/JSON export now emits "category": "ai_quality" for PHR items where consumers previously saw "completeness" (asked about this in #473 — happy to add a changelog note if wanted).

Validation

• pytest — 6682 passed. Excluded on my machine only: TestCalibrateRunWriteGuard (click 8.2 removed mix_stderr), test_date_parsing (timezone-dependent assertion, passes under UTC=+0 only), test_model_consistency_check_passes (SECURITY.md missing 2.51.x, pre-existing on main), and one rich-output snapshot test — all verified failing on a clean checkout of main as well
• New regression tests verified to fail against unpatched main and pass with this change
• ruff check src/ tests/ — clean
• mypy src/drift — no new errors (same 2 pre-existing cli.py errors as main)

Closes #473

[7arnav1]

I don't have permission to add labels — this should be release:fix per the labeling convention.