feat: expose runner logs through CLI and agent tooling

[forestileao]

Problem

Runner failures are hard to debug because runner/broker logs are not easily available through product tooling. This blocks investigation when runners are flaky or fail to start.

Proposal

Expose runner logs through both CLI and agent tooling.

Acceptance criteria

• CLI can fetch relevant runner logs for a run/node/execution.
• Agent tooling can fetch the same logs in a scoped way.
• The output includes enough metadata to correlate logs with the failing runner/execution.