[feat] Node observability: metrics, probes, and structured logs

[St4NNi]

Closes #278

Gives a node the surfaces an operator needs to run it.

• Per-node Prometheus metrics registry with text exposition, exported on a dedicated ops listener alongside /healthz and /readyz. The ops server starts early and gates readiness during node bootstrap, so a node does not advertise itself as ready before it can serve.
• REST and S3 request metrics recorded in middleware, with resolved operation labels and nonstandard HTTP methods collapsed into a single other label so the label space stays bounded.
• Storage and queue gauges exported on scrape, with queue lag exposed as reusable probe snapshots. Storage in-flight is balanced by a drop guard so a cancelled request cannot leak the gauge.
• Optional JSON log format for structured log ingestion.