feat: wire AgnoOS monitoring for observability

[Fieldnote-Echo]

Summary

Grippy currently has no structured observability beyond stdlib logging. Agno ships AgnoOS which provides instant dashboards for agent runs — token usage, tool call traces, latency, and more — with minimal wiring.

Motivation

Grippy's LOW review findings on PR #30 identified three observability gaps:

• Search fallback is silent (can't distinguish "no results" vs "backend error")
• No index build stats (per-extension counts, timing, chunk sizes)
• No cache invalidation visibility (addressed in 2d3c7f7 with log.info)

All three are "add a log line" fixes, but without a dashboard nobody reads them. AgnoOS solves this at the agent level where the interesting telemetry lives.

Proposed approach

1. AgnoOS for agent telemetry — monitoring=True on the Agent() constructor + AGNO_API_KEY env var. Gives users instant dashboards for:

   • Run traces per review (tokens, latency, tool call chains)
   • Search hit/miss rates
   • Structured output validation/retry patterns

2. stdlib logging stays for non-agent code — codebase indexing, rule engine, diff parsing are deterministic and fast. These run in throwaway Actions containers where structured logging (structlog, etc.) adds complexity without value.

3. Optional — users add AGNO_API_KEY to their repo secrets and get observability for free. No key = no monitoring, zero behavior change.

What this is NOT

• Not adding structlog or custom telemetry plumbing
• Not adding metrics endpoints or Prometheus exporters
• Not changing log format for CI output

References

• PR feat: migrate CodebaseIndex to Agno LanceDb with hybrid search #30 Grippy review findings (LOW observability items)
• AgnoOS docs
• Commit 2d3c7f7 (cache invalidation logging)