test(core): wait out ETXTBSY before running the fake container CLI

[vieiralucas]

Summary

container_image::tests::a_missing_image_fails_without_retrying failed in CI on #2521 with:

left:  Err("/tmp/.tmpcxArB6/cli pull: Text file busy (os error 26)")
right: Err("Error response from daemon: manifest for alpine:nope not found: manifest unknown")

The FakeCli fixture added in #2516 writes a shell script and then executes it. Tests run on parallel threads. A child process forked by another test while fs::write still had the script open inherits that write descriptor, and holds it until the child calls exec. If our test executes the script in that window, it fails with ETXTBSY. It's a race in the test fixture; the production pull code is fine.

Change

Right after writing the script, FakeCli::new runs it once and retries (5ms apart, up to 1s) while that fails with ErrorKind::ExecutableFileBusy. Once one run succeeds, no process can hold a writable descriptor to the script again, so the real calls can't hit the race. The probe's line in calls.log is removed so call assertions are unchanged.

Test plan

• cargo test -p fakecloud-core container_image: 17 pass
• cargo clippy -p fakecloud-core --all-targets -- -D warnings: clean

No non-code surface affected.

---

Summary by cubic

Fixes a flaky failure in the fake container CLI test fixture where a test could fail with Text file busy (os error 26) when another parallel test's child process still held the script open for writing.

FakeCli::new now runs the script once right after writing it, retrying while it gets ExecutableFileBusy, guaranteeing no process can later hold a writable descriptor. The probe run's line is removed from calls.log so call assertions are unchanged.

^{Written for commit dbe6565. Summary will update on new commits.}