[Priyanshu-u07]

#442 gives every inference request an X-Request-ID, but the stored log row never kept it, so a request seen in a client's error or a log line could not be found in the logs.

• Add a request_id column and index to inference_logs: global_schema.sql for fresh installs, migration 20261026_inference_logs_request_id.sql for existing ones (idempotent; the index is built CONCURRENTLY after an SPLIT marker, as in 20260530_pool_first_deployments.sql, so it doesn't block inserts).
• The inference client sends the request's id with each log; the gateway stores it.
• Return request_id with each log, and let GET /insights/logs filter by it (?request_id=).

Rows logged before this keep request_id null.

Tests: the id is stored (and null without one), the insights filter uses it, the client sends the id of the request being logged. Applied with the repo's migration runner twice: the index is built concurrently and valid, and the second run is a no-op. Full suite with the CI ignore list: 3,073 passed.

Follow-up: show the request ID in the Insights log details and add a search box next to the IP filter.