Improve symdb probe-locations reliability and output

[ojung]

Summary

Improve symdb search reliability for large services by adding retry logic, partial-result polling driven by indexing_status, and richer probe-location output with method signatures.

Changes

• Add retry logic with exponential backoff for transient 502/503/504 errors (src/client.rs, src/commands/symdb.rs)
• Add HttpError struct preserving status codes for programmatic retry decisions (src/client.rs)
• Request partial results (allow_partial=true) and poll until every service-version reports a terminal indexing_status (COMPLETED, SOME_FAILED, ALL_FAILED, NO_ATTACHMENTS) with a 60s deadline (src/commands/symdb.rs)
• Add --no-allow-partial flag to disable partial results when callers prefer waiting for full indexing (src/main.rs, src/commands/symdb.rs)
• Switch fetch_children_bulk from /scopes/children?scope_name=... to /scopes/{id}/children (src/commands/symdb.rs)
• Add format_probe_location helper that extracts ARG-type symbols into type:method(arg1, arg2, ...) matching the API's where.signature format (src/commands/symdb.rs)
• Track seen locations and class IDs across polls via ProbeCollector to avoid duplicate children fetches (src/commands/symdb.rs)
• Add signature field to ResolvedProbe and pass it through to where.signature in probe creation payload (src/commands/debugger.rs)
• Both TYPE:METHOD and TYPE:METHOD(args) accepted as --probe-location values (src/commands/debugger.rs)
• Update skill docs and docs/EXAMPLES.md to document the new signature format

Testing

Tested against large Java services (500–2600 probe locations). All complete within the 60s deadline, typically 12–29s depending on service size and indexing state.