Feat/benchmark job lifecycle

[amh1k]

Summary

Implement the benchmark Job lifecycle through Coordinator.Run: create temporary credentials and a runner Job, wait for execution, collect output, and clean up Kubernetes resources.

Closes #17

Changes

• Integrate Job and Secret creation into Coordinator.Run.
• Poll Job completion and failure conditions.
• Discover the runner Pod and verify its Job ownership.
• Collect up to 1 MiB of runner output and report truncation.
• Preserve available logs on failure, with a separate diagnostic timeout.
• Enforce the configured execution timeout and propagate caller cancellation.
• Delete Jobs using UID preconditions and foreground propagation, then wait for deletion confirmation.
• Reserve time for credential Secret deletion even when Job cleanup times out.
• Preserve execution, log retrieval, and cleanup errors together.

Testing

• Added tests for Job polling, Pod ownership, log collection, and output truncation.
• Added lifecycle tests for failed Jobs, cancellation, execution timeouts, caller deadlines, and missing Pods.
• Added cleanup tests for deletion order, delayed deletion, and timeout handling.
• Passed backend tests, go vet, and git diff --check.

Scope and limitations

This PR provides the internal coordinator execution flow. Benchmark creation and status API endpoints are covered by #12.

Output is returned as raw text; metric parsing and persistence are separate work. Credential sanitization remains outstanding. A coordinator crash can leave temporary resources behind; automatic recovery is not implemented.