Improve --json catalog for LLM consumption

[gmoon]

Summary

• Add concepts block with node types, edge semantics, versioning model, and ID conventions so LLMs understand the domain model before using commands
• Add workflows section with 4 task-oriented patterns (capture_decision, check_health, respond_to_drift, record_gap)
• Improve all command descriptions with "when to use" guidance
• Add output_schema_hint per command so LLMs know response shapes without speculative execution
• Convert examples from bare strings to {command, explanation} objects
• Add related_commands for navigation between commands
• 6 new tests validating concepts, workflows, structured examples, schema hints, related commands; edge type test validates against canonical EDGE_TYPES constant

Closes #22

Test plan

• make pre-commit passes (fmt, clippy, tests)
• cargo run -- --json | jq .concepts returns node types and edge semantics
• cargo run -- --json | jq .workflows returns 4 workflows
• cargo run -- --json | jq '.commands[0].examples[0].command' returns structured example
• All 9 catalog tests pass including 6 new ones

🤖 Generated with Claude Code