Treat purpose-oriented questions as overview cues so ask responds with a workspace brief instead of a findings dump.

When using --llm, add bounded local code excerpts (routes/models/httpx) so the model can answer API-surface questions and follow-ups precisely.

Expose enumerate results (e.g. missing upstream routes) to the LLM so --llm can answer coverage questions precisely.

Previously http_call edges only sent edge_type without the method,
path, or URL. This meant the API couldn't match inbound calls to
specific routes, falling back to the coarse remote_server reference.

Now EdgeRecord includes http_method, http_path, and url fields for
http_call edges, enabling precise route-level inbound caller tracking.

URLs like 'localhost:8081/orders' (without http://) were not having
their path extracted. Now we handle both formats.

When --verbose is set, print inbound_route_dependents keys from the API and rendered route keys to diagnose missing upstream callers.

Update changelog and bump crate version.