perf(agent): reuse observed pod for non-DRA restores

[hutm]

Problem

The restore controller already has the scheduled Pod from its informer, but GPU discovery performs another Pod GET before deciding whether DRA is relevant.

Why

For Pods without resource claims, the immutable Pod spec already proves that DRA lookup is unnecessary. Removing the redundant GET reduces API traffic and avoids an extra request on the restore path.

Change

Reuse the observed Pod when its identity matches and it has no resource claims. Pods that may use DRA retain the live Pod lookup because generated claim names arrive through Pod status. GPU assignment still comes from kubelet PodResources.

Local phase measurements showed a smaller GPU-discovery preflight, but no statistically significant end-to-end latency improvement.

Testing

• Added coverage verifying that a matching Pod without resource claims performs no Pod GET.
• Ran the complete agent test suite.

Fixes #352

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

_{Comment @coderabbitai help to get the list of available commands.}