test: gate st2 against bounded PTY fleet observation

[schickling-assistant]

Summary

• add a hermetic packaged st2 -> pty fleet-observation contract for 0/75/100/500 sessions
• pin the exact PTY producer revision and execute its shipped fleet-observation fault suite for EPERM handling and one fleet-wide socket-fallback budget
• preserve the existing 2-second st2 outer deadline and reject malformed partial snapshots atomically

This is the producer/consumer evidence replacement for the synthetic fixed-sleep fixture in #78. Its scope is deliberately limited to the exact packaged pty list --json producer and the fleet-observation behavior exercised here; it does not claim the broader runtime lifecycle contract tracked in #30. It also does not increase the timeout or ratify full-registry census as the final architecture. The follow-up architecture is pure exact-ID observation on reconciliation paths, with whole-root snapshots reserved for explicit audit/eval cleanup.

Verification at e1e06669665f8746cb767960ebbb0632770fcf1b

• hosted nix flake check --print-build-logs: PASS
  • run: https://github.com/compoundingtech/st2/actions/runs/30547906736
  • job: https://github.com/compoundingtech/st2/actions/runs/30547906736/job/90888710821
  • includes the dedicated one-test atomic snapshot target and the packaged PTY fleet-observation check
• cargo test --test atomic_pty_snapshot -- --test-threads=1: PASS (1/1)
• cargo test --workspace --all-targets -- --test-threads=1: PASS
• cargo clippy --workspace --all-targets -- -D warnings: PASS
• cargo doc --workspace --no-deps: PASS (five unchanged pre-existing warnings)
• git diff --check origin/main: PASS

Review corrections

• narrowed the gate and PR wording to the exact packaged fleet-observation contract while Activation: deploy PTY environment persistence for manual restart #30 remains open
• replaced newly introduced PTY seat fixture names/comments with session terminology

Named follow-up

Export the PTY producer fault contract as a stable public flake check or test entrypoint. The current hermetic compatibility gate intentionally executes the exact pinned package, but reaches its private lib/pty/node_modules/vitest layout until PTY publishes that seam.

No live catalog, service, or PTY state was used.

[myobie]

The producer/consumer observation proof is useful and the test shape is sound, but two corrections are required before approval:

1. Narrow the flake wording from broad “runtime compatibility” / “coherent lifecycle baseline” to the exact contract this proves: packaged PTY fleet-observation compatibility. st2 issue #30 documents a still-open packaged restart-environment incompatibility, so this pin cannot honestly stand for full st2↔PTY runtime compatibility.
2. Replace the newly introduced seat terminology (500-seat, variable name, and fixture prefix) with session terminology. These are PTY sessions, and st2’s accepted ontology models agents rather than seats.

No objection to the evidence architecture once those scope/terminology corrections are made.

[myobie]

Approved at exact head e1e0666. Both requested corrections are complete: the pin is scoped to packaged fleet observation rather than broad runtime compatibility, and all newly introduced PTY fixtures use session terminology. The producer/consumer proof, atomic malformed-snapshot refusal, fleet-size budget matrix, focused/full checks, and hosted Nix gate are green.