[Spicy-cream]

What problem does this PR solve?

URMA needs connection-level statistics and resource watermarks to locate SQ pressure, remote receive-credit starvation and registered-buffer exhaustion, and to relate transport waits to individual RPCs.

What is changed and the side effects?

Changed:

• Add process-wide bvars and per-connection statistics for bulk writes, data WRs, resource waits, errors and handshake outcomes.
• Export registered BufferPool occupancy and watermarks through JSON socket snapshots.
• Add sampled RPC transport tracing and a diagnostic CLI for queries and rate watching.
• Keep CQ processing runnable for traced responses, reclaim send slots after their buffers, and publish mock send completion after consuming source SGEs.

Transport tracing is disabled by default and configured with urma_trace_sample_every.

Check List:

• URMA mock and TCP-only builds pass.
• Existing URMA, Socket and builtin-service suites pass; new observability and CLI tests pass.
• Event-mode and polling-mode data-path tests each pass 50 consecutive runs.