fix(evolution): honor disabled planning rules in planner

[cursor]

Summary

• skilllite evolution disable <rule_id> wrote "disabled": true and reported success, but PlanningRule had no disabled field and the planner kept injecting the rule into future plans/beliefs.
• Persist PlanningRule.disabled, filter disabled rules from planner/beliefs consumption paths (not from seed merge loaders, to avoid write-back data loss), and add regression coverage.

Trigger scenario

1. An evolved mutable rule exists in chat/prompts/rules.json.
2. User runs skilllite evolution disable <rule_id> and sees success.
3. Before this fix, the next chat/plan still loaded and injected that rule because serde dropped disabled and no consumer filtered it.

Task Linkage

• Task ID: TASK-2026-072
• Task folder: tasks/TASK-2026-072-evolution-disable-honored/
• Contribution mode: core/maintainer (strict)

Injected Specs

• spec/verification-integrity.md
• spec/task-artifact-language.md
• spec/architecture-boundaries.md
• spec/structured-signal-first.md
• spec/capability-gap-evolution.md
• spec/rust-conventions.md
• spec/testing-policy.md
• spec/docs-sync.md (N/A content change; corrects existing disable semantics)

Validation Evidence

• Commands executed:
  • cargo fmt --check
  • cargo test -p skilllite-core planning
  • cargo test -p skilllite-agent disabled
  • cargo test -p skilllite-core -p skilllite-agent -p skilllite-evolution --lib
  • cargo clippy -p skilllite-core -p skilllite-agent -p skilllite-evolution --all-targets -- -D warnings
  • python3 scripts/validate_tasks.py
• Key results:
  • planning serde tests: 2 passed
  • agent disable-filter tests: passed
  • touched-crate lib tests: passed
  • clippy clean under -D warnings
  • task validation: 71 task directories checked

Docs Sync

• EN changed: none
• ZH mirrored: N/A
• N/A reason: no command/env/docs surface change; behavior now matches the existing disable command contract

Regression Scope

• Areas likely affected:
  • Planning rule deserialize/serialize
  • Task planner rule injection
  • Beliefs block derived from evolved rules
• Explicit non-goals / follow-ups:
  • --workspace for disable/explain
  • Separate sweep findings: skill-add path traversal via SKILL.md name; MCP skill_name join without single-segment validation

Checklist

• cargo fmt --check
• Targeted clippy on changed crates with -D warnings
• Relevant tests added/updated and passing
• Docs sync N/A with reason
• No secrets committed