Smoke test: run nightshift dry-run + basic cycle as post-merge gate

[fazxes]

Problem

Even between full E2E evaluations, there is no quick sanity check that the nightshift CLI actually works. Unit tests pass but the CLI could be broken (import errors, missing config, bad shell commands). We need a fast smoke test after every merge.

What needs to happen

1. Add a post-merge smoke test to evolve.md Step 9 (post-merge health check). After make check passes:

   • Run python3 -m nightshift run --dry-run --agent claude (already in Step 5 but often skipped)
   • Run python3 -m nightshift run --dry-run --agent codex
   • Both must exit 0

2. Update evolve-auto.md — add SMOKE TEST RULE: "Dry-run is mandatory post-merge, not optional."

3. Update scripts/smoke-test.sh if needed — ensure it can run headless without interactive prompts.

4. Update docs as needed.

Acceptance Criteria

• Post-merge step runs dry-run for both agents
• Failure blocks the session from reporting success
• Agent cannot skip this with "dry-run skipped" rationalization
• Docs updated

[fazxes]

Converted to task #93