feat(workflow): dispatch durable lanes without root model

[Hmbown]

Summary

• dispatch codewhale workflow run directly through the host-owned Workflow tool, without an operator-model turn
• preserve profile/provider/model, approval, sandbox, network, MCP, environment, and keyring precedence across the hidden runner
• restore named Fleet role resolution and persist typed Workflow receipts with distinct Fleet and Workflow run IDs
• harden Lane tmux execution with a Rust log proxy, bounded atomic receipts, per-Lane lifecycle locks, pinned tmux sockets, and fail-closed backend behavior

Release issues

• Advances v0.8.68 Phase 2: Workflow steps reference Fleet roles (not raw prompts/profiles) #4177
• Advances v0.8.68: Stopship workflow as fleet-backed lane (dogfood #4090/#4093/#4094) #4178
• Advances v0.8.68 Phase 3: Workflow gates and handoffs between Fleet roles #4179

Verification

• cargo test -p codewhale-protocol --quiet: 43 unit + 13 parity passed
• cargo test -p codewhale-workflow --quiet: 102 passed
• cargo test -p codewhale-lane --quiet: 17 passed
• cargo test -p codewhale-cli --quiet: 130 library + 1 shim passed
• cargo test -p codewhale-tui --test workflow_tool_stream_acceptance --quiet: 3 passed
• full default-feature codewhale-tui suite with only GROK_HOME isolated: 6,310 passed, 0 failed, 2 ignored
• GROK_HOME-isolated cargo test --workspace --all-features --locked --quiet: all workspace, acceptance, and doctest targets passed; the all-feature TUI binary reported 6,325 passed, 0 failed, 2 ignored
• exact release-workflow all-target/all-feature Clippy command, formatting, diff hygiene, debug CLI/TUI build, and scripts/release/check-versions.sh passed
• cargo check -p codewhale-lane --target x86_64-pc-windows-gnu --locked passed
• cargo check -p codewhale-lane --target aarch64-linux-android --locked passed
• live tmux proofs covered binary and unterminated output, 200 KB newline-free framing, failed-kill rollback, terminal reconciliation, missing tmux, VM fail-closed behavior, and TMUX_TMPDIR drift
• final GitHub matrix on e021d2cae: 18 successful, 4 intentionally skipped, 0 failed/pending; Windows passed in 9m59s and macOS plus offline eval passed in 11m21s

The ambient full TUI run inherited a live Grok OAuth catalog and reproduced seven model-picker fixture failures on the clean base. Isolating only GROK_HOME preserved the normal Codewhale/Codex catalogs and produced the clean results above. The local host lacks x86_64-w64-mingw32-gcc, but the complete GitHub Windows workspace run is green.

Release boundary

• no paid model or inference calls were made
• scripts/release/check-published.sh 0.8.68 truthfully reports that npm and the release crates are not published yet
• this draft does not merge, tag, publish, deploy, close issues, or create paid resources
• VM and CI Runtime backends remain explicit terminal failures until implemented

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[claude]

Claude finished @Hmbown's task in 2m 31s —— View job

---

Review in progress

• Gather context (PR body, changed files, CLAUDE.md/AGENTS.md guidance)
• Fetch full diff (gh pr diff 4336, base ae1dcd24..head e021d2cae)
• Review crates/lane/src/runtime.rs + registry.rs (tmux lifecycle, log proxy, fd-lock, receipts)
• Review crates/cli/src/lib.rs (workflow dispatch, env/keyring precedence)
• Review TUI/protocol changes (main.rs, tools/workflow.rs, config.rs, fleet/*, protocol/fleet.rs, subagent + acceptance tests)
• Check provider/model prefix inference safety (v0.8.65 invariant)
• Synthesize findings and post final review