Always list all container states so finished jobs are visible

[Shine-neko]

Problem

build_list_options set the Docker list_containers all flag conditionally:

let all = !matches!(status, "all");

For a status = "all" lookup this yields all = false, so Docker returns running containers only. Every exited container is invisible to that listing — the opposite of what the function's own comment promises ("Always list every container and filter by state client-side").

This breaks the job path. handle_job_deployment calls list_instances(deployment.id, "all") to find its container and converge to Completed. A pg_dump backup job (kind: job, replicas: 1) exits ~0.5s after starting; on the next scheduler tick there are zero running containers, so the inverted flag makes the listing empty even though the finished container is right there with a matching ring_deployment label. Ring concludes the job has no instance, recreates a new container, and sets Running — every tick, forever.

In production this looped ~30 backup deployments and piled up ~8900 Exited(0) postgres:18-alpine containers, saturating the reconcile cycle (which slowed everything else, including unrelated redeploys).

Regression from a Podman fix (605a25c, 2026-06-19) that correctly stopped a different loop but tied all to the filter.

Fix

Make the flag unconditional:

ListContainersOptionsBuilder::new().all(true).build()

Listing is exhaustive; matches_status already does per-state filtering client-side (including "all", "active", "exited", ...), so this restores the intended behaviour without reintroducing the Podman created-container regression (that fix lives in matches_status's active arm, untouched here).

Verification

• New regression test list_options_always_request_all_states asserts all == true for every filter. Confirmed it fails when the buggy expression is restored and passes with the fix.
• Deployed to production: container recreation stopped immediately (0 recreations post-deploy vs ~30/cycle before), and the Exited-container count stopped growing.
• Full suite green, cargo fmt/clippy clean.

Follow-ups (out of scope): reap accumulated exited job containers; consider making list_instances("all").first() pick the most-recent container explicitly rather than relying on Docker list order.