feat(resume): carry a session across models and machines

[AetherAI3]

The one capability to remember

Start a task on one model. Finish it on another, on another machine. Your tests decide when it's done.

aether resume could already replay a prior session's transcript to the screen — but the brain never saw a byte of it. Continuing meant re-typing the story so far, and only ever on the machine where the work started. This PR makes the resume path mean what it says.

What's in it

A handoff (src/core/handoff.ts) is the machine-facing half of a session log: a small JSON document distilled from one run — the task, the model that ran it, the verify gate's verdict, the failing-test count, the files it changed, the verification command, and the repository identity (origin remote, branch, HEAD).

• aether agent --resume <id> prepends a continuation brief built from that record to the task the brain receives, so a different model picks the thread up with the project context in hand. With no new task the run continues the original one — it used to fail with nothing to do.
• aether resume export [id] [--out <file>] writes the handoff to a file. Nothing in it is keyed to an absolute path, so it can be copied to another checkout, machine, or OS, and aether agent --resume <file> continues there. A file reference is deliberately not workspace-scoped; a session id still is.
• It is a summary, not a transcript: no file contents, no shell commands, no credential-shaped values. Untrusted files are validated field by field on the way back in, and one written by a newer Agent is refused with an upgrade hint rather than half-read.

Three bugs this needed, each real on its own

aether agent --local "<task>" was dead out of the box	It spawned the separately-installed Python brain unconditionally, so a plain npm i -g aether-agents could only answer spawn python ENOENT. The one-shot offline path now drives the Ollama brain that ships in the package — the same one the REPL's --local turns already used. AETHER_LOCAL_BRAIN=python opts back in; the choice is a pure decision beside chooseBackend.
Session logs redacted every file path	The credential filter matched pat as a substring, so path (and PATH, patch, pattern) was stored as [REDACTED]. Every log could say a file changed but not which one. pat is now anchored to a whole segment; pat, gh_pat, pat_token are still redacted.
A run that never reached your tests claimed a failing one	The Ollama brain reported a placeholder remaining: 1, so an unreachable-Ollama turn printed 1 test failing when nothing had run. The count now comes only from the host's own verify run.

Proof

npm run demo:handoff     # ~5s, no account, no model download

scripts/handoff-demo.ts builds a throwaway git repo with two genuinely failing tests, runs the real CLI on model A until half the work is done, exports the handoff, creates a second checkout at a different absolute path, deletes the first checkout and its session logs, then finishes the job on model B with --resume <file> and no restated task.

The model — and only the model — is a scripted local HTTP server speaking Ollama's OpenAI-compatible endpoint, which is what makes the run deterministic enough to be a gate. Everything else is real: the CLI, the git repos, the file edits, the permission gate, the handoff file, node --test, and the verify gate. It asserts three things independently:

1. session B's first prompt contained the continuation brief, naming model A and src/slug.js;
2. node --test is green when run by the script, outside the agent;
3. the CLI's own verify gate exited 0.

AETHER_DEMO_REAL=1 runs the identical script against real Ollama models. Documented in docs/demo/handoff.md, including the 20–45 second recording sequence.

Sample run:

── The handoff ───────────────────────────────────────────────────────
⇄ handoff written .../handoff.json
  session 2026-08-19T…-local-22556 · incomplete · 1 file(s) · 2 step(s)
   carried: model aether-demo-a, status incomplete, files ["src/slug.js"],
            repo https://github.com/aether-demo/slugify.git

── Machine B — aether-demo-b ─────────────────────────────────────────
   no task restated; the handoff file is the only context
  ✓ ok · 1 file changed · tests green · 1s

── PASSED ────────────────────────────────────────────────────────────
   One model started it. Another finished it, elsewhere, with no re-pasted context.
   The tests decided when it was done.

Verification

• npm run typecheck — clean
• 65 unit tests across handoff, resume_cmd, backend_select, session_log, session_resume, cli_registry, command_docs_parity, brain_ollama, ui, version — green locally
• npm run demo:handoff — PASSED
• npm run smoke — 3 pass / 4 skip / 0 fail (skips need Ollama, network, or sign-in)
• npm pack --dry-run — clean; the production verifier's pack-install-launch test is green
• Failure paths exercised by hand: missing Ollama, a corrupt handoff, a missing handoff file, and an unknown session id all fail with an actionable line

Full-suite CI is the gate here — one tool_executor test is pathologically slow on the authoring machine and was not run locally.

Notes for review

• src/core/brain_ollama.ts is also touched by fix(brain): correlate Ollama tool results by id, generate schemas, stop faking steer [SC-A4.1] #75; the overlap is the single remaining line.
• Stacked: release/v0.2.0 carries the README hero, release notes, and the version bump, and should merge after this.