e2e: cover podman logs and metrics

[Shine-neko]

Podman goes from 5 to 6 e2e tests, covering the two things that make a runtime observable: logs and metrics.

What was added

t6_podman_logs_metrics.sh — ring deployment logs returns the container's stdout, --tail N actually bounds the output, and ring deployment metrics reports real per-instance figures.

Podman shares Docker's lifecycle code but not its daemon: logs and stats travel over Podman's Docker-compatible API, rootless, through a different socket. "It works on Docker" says nothing about whether that compatibility layer returns what Ring expects — and a runtime whose logs or metrics come back empty is effectively unobservable.

The memory assertion checks both that usage is non-zero (a real cgroup read, not a placeholder) and that the manifest's 64Mi limit is reflected back — which proves the limit reached the container and that Ring reads it from the runtime rather than echoing the manifest. Observed: 400.00 KiB / 64.00 MiB (0.6%).

A trap worth knowing about

Only three deployment subcommands accept --output json:

supports --output json	renders text
list, inspect, health-checks	metrics, events, logs

Passing --output json to a text command is not always a hard error, and piping rendered text through jq yields nothing silently — which reads exactly like "the feature is broken". That mis-diagnosis cost me two false conclusions in this suite: first that Firecracker's on_failure alert did not fire (it does, fixed in #213), then that Podman reported no metrics at all (it does).

Documented at the top of tests/e2e/lib.sh so the next person writing assertions does not repeat it.

Verification

podman t1–t6 all run and pass locally against a real rootless Podman socket.