Retry instance directory deletion on ENOTEMPTY

[sjmiller609]

Summary

• retry instance directory deletion when os.RemoveAll loses a race with concurrent writers and returns ENOTEMPTY
• use exponential backoff starting at 10ms and capping at 100ms for up to 5 retries
• add unit coverage for retry success and retry exhaustion

Testing

• go test ./lib/instances -run 'TestRemoveAllWithRetry|TestMetadataPersistence' (fails in this checkout because embedded binaries lib/system/guest_agent/guest-agent and lib/hypervisor/vz/vz-shim/vz-shim are missing)

---

Note

Medium Risk
Touches instance deletion cleanup by adding retry/backoff around os.RemoveAll, which could impact teardown timing and error handling if the retry logic misbehaves. Change is localized and guarded to only retry ENOTEMPTY with a small capped backoff.

Overview
Improves instance teardown robustness by retrying instance directory removal when os.RemoveAll fails with ENOTEMPTY, using an exponential backoff (10ms doubling, capped at 100ms) for up to 5 retries via removeAllWithRetry.

Adds unit tests covering both the successful retry path and the max-retry exhaustion behavior, including verification of the backoff schedule.

^{Reviewed by Cursor Bugbot for commit 4a33f36. Bugbot is set up for automated code reviews on this repo. Configure here.}