Flaky: TestDockerExecInteractive fails on first CI run (exit 126), passes on re-run

[dpup]

Summary

TestDockerExecInteractive (in internal/container) is flaky in CI: it fails on the initial run of nearly every push and passes on re-run, with no change to the exec code path between attempts.

Failing subtests

• TestDockerExecInteractive/TTY_output_contains_expected_string
• TestDockerExecInteractive/exit_code_propagation_via_ExecError

Symptom

exec_interactive_test.go:54: ExecInteractive failed: exec failed with exit code 126
--- FAIL: TestDockerExecInteractive (10.21s)
FAIL	github.com/majorcontext/moat/internal/container

Exit code 126 is "command found but cannot execute" — consistent with a TTY/exec-setup race in the Docker-in-Docker CI environment (e.g. the container/exec not being fully ready when ExecInteractive runs), rather than a logic bug.

Frequency / evidence

On PR #404 the Test job failed on the first attempt of four consecutive pushes and passed on every manual re-run:

• run 28187501415 (failed → re-run passed)
• run 28186749229 (failed → re-run passed)
• run 28138719182 (failed → re-run passed)

Not reproducible locally: the full internal/container race suite (go test -race ./internal/container/...) passes repeatedly on a local Docker-in-Docker setup.

Impact

• Blocks the first CI Test run on essentially every push; requires a manual "re-run failed jobs".
• Wastes CI minutes and erodes signal — a real failure in this job is easy to dismiss as "the usual flake".

Suggested fixes (pick one)

1. Stabilize: wait for container/exec readiness before the interactive exec in the test (poll until the shell is execable), or add a bounded retry around the 126-class transient.
2. Quarantine: tag the test (build tag or t.Skip under a MOAT_FLAKY=skip env / testing.Short()) until (1) lands, so CI stops false-failing.

Surfaced while landing #404 (volume-mode workspaces); this test is pre-existing and unrelated to that change.