Add status subcommand: read-only deployment view across hosts

[jfding]

Summary

Adds git-supervisor status — a read-only subcommand that SSHes each configured host in parallel and prints a grouped per-host/per-repo view of deployed branches, releases, the prod.latest symlink target, stale dirs, and user dot-flag files. Read-only: no fetch, no clone, no docker actions.

Example output:

host: app1
  webapp  (latest: v2.1.5)
    dev          def4567  3d ago      debugging
    main         abc1234  2m ago      -
    --
    v2.1.5       fed6543  1h ago      -
    v2.1.4       cba9876  2026-04-12  -
    stale:
      webapp.oldfeature.to-be-removed   3d ago

host: app3  ERROR
  ssh: connect to host app3.example.com port 22: Connection timed out

What's in this branch

• New subcommand Status(StatusArgs) in src/main.rs, with --host PATTERN (repeatable, glob */?, union semantics; zero matches exits non-zero).
• Remote probe src/status_probe.sh (~120 lines bash) embedded via include_str!. Uses the same DIR_BASE env contract as check-push.sh. Longest-prefix matching against \$DIR_REPOS correctly handles dotted repo names like my.api. Distinguishes missing \$DIR_COPIES (legitimate empty, exit 0) from unreadable (stderr + exit 1).
• Renderer in src/status.rs: grouped hierarchical layout, branches sorted alphabetically, releases sorted version-desc via a Rust port of check-push.sh's sort_version_tags_desc (handles vX.Y.Z and vYYYYQN.X.Y Q-notation). Adaptive relative time (just now / Nm/Nh/Nd ago / YYYY-MM-DD after 7d). 7-char SHA truncation. Color rules for .debugging/.stopping (red), .skipping (grey), stale section (yellow), latest annotation (green/yellow/red by health).
• Parallel fanout via std::thread::scope, mirroring run_cycle. Output buffered and sorted by host ID for deterministic rendering.
• SSH refactor in src/ssh.rs: extracted build_ssh_command shared by ssh_run / ssh_run_with_stdin / new ssh_run_capture. ssh_run_capture pipes all three streams and uses wait_with_output to drain stdout+stderr concurrently (avoids a deadlock latent in the naive sequential-read pattern).
• Failed/empty hosts appear in the table as ERROR (red) or (no deployments yet) (grey) — never silently dropped.

Test plan

• Unit tests: 24 in src/status.rs (parser + helpers: time format, glob, version sort, group, repo header)
• Unit tests: 5 in src/ssh.rs (incl. ssh_run_capture stdout + stderr+exit-code propagation)
• Unit tests: 2 in src/main.rs (CLI parser for --host flag)
• Integration tests: 3 in tests/integration_status.rs against localhost (happy path, dotted repo my.api, --host zero-match)
• Full suite: `cargo test` → 79 passed, 2 ignored (no regressions in existing 47 tests)
• `cargo clippy --all-targets -- -D warnings` → clean
• `cargo build --release` → clean

Design plan

The design was iterated through a structured interview before implementation (see `docs/superpowers/plans/2026-05-20-01-status-subcommand.md` for the full rationale on each decision: output format, layout, sort order, repo-name parsing, concurrency, SSH helper architecture, exit code semantics, time format, probe robustness, env-var contract, filter flags, renderer specifics, and test tiers).

🤖 Generated with Claude Code