CI: quiet transient multi-CN startup probes

[XuPeng-SH]

Root cause

The standalone multi-CN workflow starts the proxy, then immediately runs an SQL readiness probe. Before a CN is routable, the probe returns "no available CN server". The existing loop intentionally retries, but piping every attempt through tee /dev/stderr surfaced normal startup retries as repeated ERROR lines. In run #28794, the first probe failed at 01:40:34 UTC, CN startup/HAKeeper registration activity followed, the last observed route failure was at 01:40:43, and the SQL probe succeeded at 01:40:44. Run #28800 shows the same transient pattern. The cited BVT failures were separate assertions.

The INFO artifacts do not reveal why the route candidate set remains empty after registration becomes visible (stale proxy snapshot versus admission, work-state, or selector filtering). This change does not alter MatrixOne routing or proxy error semantics.

Change

• Capture retry output instead of printing every failed probe.
• Keep the existing bound of 300 attempts and one-second inter-attempt delay.
• Report elapsed time and successful attempt; if all attempts fail, print the last probe output plus Compose status and recent logs, then fail.

Validation

• Workflow YAML parsed successfully.
• Readiness step passes bash -n.
• In-memory success and timeout-path mocks passed.
• Live multi-CN Compose was not run locally.

Fixes #450