FE-1107: Port cook orchestration into code/executor tools

[kostandinang]

Stack Context

FE-1107 closes executor surface residue after the KA executor lane landed the real execute_* run machinery. This PR is stacked after the executor run integrity/observability work and reconciles mode vocabulary, Pi adapter topology, prompt conduct, and canonical docs.

What?

• Renames the live Specify-mode runtime id from elicit to specify while preserving the elicitor role and elicit skill name.
• Moves executor execute_* Pi adapters and passive run-update observers from agent-runtime/ to src/.pi/extensions/executor/.
• Retires live orchestrator_stub prompt/test/doc witnesses in favor of the active execute_* surface and explicit host-promotion acceptance boundary.
• Pins executor observer contracts: tolerant run artifact reads, populated-plan preference, active stream tails, and stream corruption tolerance.
• Updates SPEC/PLAN/topology/testing docs to match the current surface.
• Records the post-PR ln-sync cleanup: FE-1107 now points at draft PR FE-1107: Port cook orchestration into code/executor tools #298, retired stub cleanup is no longer a live Parallel item, and the remaining executor card is marked retained-unconfirmed pending KA confirmation instead of being deleted.

Why?

The execution mechanism exists, but stale mode names and stub-era executor text made the product read as though Execute mode still had no live execution boundary. This keeps operational mode, product label, foreground role, Pi adapter ownership, and executor-run observer contracts separate and explicit.

Verification

• npm run fix
• Targeted runtime/executor suite: 14 files, 135 tests
• npm run verify (214 test files passed, 1479 tests passed, build passed)
• npm run check after the sync-only follow-up commit

[kostandinang]

• FE-1107: Port cook orchestration into code/executor tools #298 👈 (View in Graphite)
• FE-1154: Harden executor run integrity and observability #297 : 1 other dependent PR (#296 )
• FE-1141: Watch executor runs live in web sidecar #295
• next

This stack of pull requests is managed by Graphite. Learn more about stacking.

[cursor]

PR Summary

Medium Risk
Wide rename of operational-mode ids and authority-matrix witnesses can break mode switch, tool gating, and run-update publishing if any call sites miss the migration; changes sit on execution-adjacent seams though verification reportedly passed full suite.

Overview
Reconciles FE-1107 executor-surface residue after the KA lane landed real run machinery: product copy and SPEC/PLAN/topology now describe Specify (specify) vs Execute (execute) instead of SPEC/CODE/elicit, and Execute is framed around live execute_* tools plus explicit-acceptance host apply—not orchestrator_stub.

Runtime and tests in the diff align witnesses to that vocabulary (mode picker/chrome, authority matrix using execute_status, executor concentricity, tool-policy messages, run-update observer path under executor/execute-run-updates).

Planning/docs mark executor prompt/topology reconciliation as largely built on draft PR #298; deferred memory/cards/executor-* cleanup stays until KA confirms. One executor card is flagged retained-unconfirmed despite checked acceptance boxes.

^{Reviewed by Cursor Bugbot for commit 12d64bf. Bugbot is set up for automated code reviews on this repo. Configure here.}