feat(engine): three-tier priority scheduling — coders before reviews (fixes #490)

[mickume]

Summary

Adds three-tier archetype-aware scheduling to _interleave_by_spec() so ready tasks are ordered: (1) auto_pre reviews, (2) coder nodes, (3) non-auto_pre review nodes. This ensures implementation work is preferred when filling pool slots, complementing the review concurrency cap from #489.

Closes #490

Changes

File	Change
agent_fox/engine/graph_sync.py	Added optional node_archetypes parameter to GraphSync and _interleave_by_spec(). Splits the regular tier into coder and review sub-tiers.
agent_fox/engine/engine.py	Passes node_archetypes from graph.nodes at both GraphSync construction sites.
tests/unit/engine/test_spec_fair_scheduling.py	10 new tests in TestThreeTierPriority (8) and TestThreeTierGraphSyncIntegration (2).

Tests

• Coders-before-reviews ordering
• Full three-tier ordering (pre-review → coders → reviews)
• Verifier placement in review tier
• Multiple review archetypes in tier 3
• Backward compatibility without archetypes
• Duration hints within tiers
• Cross-spec round-robin within coder tier
• Unknown archetype defaults to coder tier
• GraphSync integration with/without archetypes

Verification

• All existing tests pass: ✅ (2 pre-existing failures unrelated)
• New tests pass: ✅ (10/10)
• Linter / formatter: ✅
• No regressions: ✅