Add local model support, health check, runner fallback, and hook fixes

[5uck1ess]

Changes

Local model support

• Update all local configs (llamacpp, ollama, lmstudio, vllm) to gemma4-26b + qwen3.6-36b
• Create ~/.pi/agent/models.json for llama-swap discovery with correct compat flags
• Add /health command to ping local endpoints before running workflows
• Add runner fallback chain (smart -> general -> fast) when a model fails
• Fix silent step skips — runner now warns when model switch fails

Hook fixes

• Change security-patterns from blocking writes to advisory warnings
• Improve context-prune types (replace unsafe casts with named interface)
• Enhance /status to show active model profile and local endpoint health

Docs

• Update README with correct local model names and /health command
• Update install.sh with local model setup instructions

All 306 tests pass.