Integrate static analysis tools for dead code detection in quality gate

[mickume]

Problem

Dead code detection in nightshift relies solely on AI prompt analysis (nightshift/categories/builtins.py lines 14-26). The quality_gate.py category detects and runs pytest/ruff/mypy but doesn't include dead-code-specific static tools. AI analysis misses dead code that deterministic tools catch reliably.

The audit found v7 had a dead InvalidTransition error variant with no code behind it — exactly the kind of thing static tools catch.

Current behavior:

• DeadCodeCategory sends file contents to an LLM with "analyze for dead code" prompt
• No integration with vulture, ruff unused rules, or language-specific dead-code detectors
• Quality gate runs pytest, ruff, mypy but not dead-code tools
• Dead code findings are nightshift-only (hunt mode) — not checked during coder sessions

Proposed Changes

1. Add dead-code static tools to quality_gate.py:

   • For Python targets: run ruff check --select F401,F811,F841 (unused imports, redefined unused, unused variables)
   • Optionally integrate vulture for deeper analysis (unused functions, classes, attributes)
   • For Rust targets: check cargo clippy dead_code warnings
   • For Go targets: check staticcheck unused results

2. Run dead-code checks in the reviewer pipeline: after a coder session commits, run the static tool on modified files. Emit findings for any new dead code introduced.

3. Add to preflight gate: include dead-code check results as a non-blocking warning (or blocking for egregious cases like entire unused modules).

Key Files

• agent_fox/nightshift/categories/builtins.py — DeadCodeCategory (AI-only)
• agent_fox/nightshift/categories/quality_gate.py — static tool runner infrastructure
• agent_fox/nightshift/categories/base.py — BaseHuntCategory two-phase pattern
• agent_fox/engine/preflight.py — preflight quality gates

Audit Evidence

From docs/audits/code_quality_5to7.md:

• "Dead code in state manager — StateError::InvalidTransition variant exists, but transition() accepts anything" (Design smell in v7)
• Static tools would have caught this deterministically without relying on AI judgment

Impact

Low effort, medium impact. The infrastructure for running external tools already exists in quality_gate.py. Adding ruff rules is a one-line config change. This catches an entire class of issues that AI analysis misses.