refactor(turn): bind the advisory primary through the shared decision owner

[huangruiteng]

Problem

main is red on full-public-smokes:

AssertionError: turn.py has 1123 lines, above budget 1114; extract a cohesive command owner before adding more code

The managed Turn host-selection wiring took loopx/cli_commands/turn.py from 1094 to 1123 lines against a frozen baseline of 1114, and the ratchet smoke (examples/cli-command-module-size-ownership-command-modularization-smoke.py) correctly rejects that.

While finding a cohesive extraction, the underlying duplication was the right thing to fix: turn.py reimplemented the controller advisory-primary rebinding inline, and loopx/cli_commands/turn_decision.py::apply_controller_advisory_primary already owns that exact rule for managed-step. Its own docstring states the intent:

run-once and managed-step must agree on what "the current governing decision" means: read the same live status, apply the same controller advisory primary, and sign the same loopx_turn_envelope_v0. Duplicating that chain would let the two drift, so both owners build through this module.

The two copies had to stay step-identical by inspection: resolve the advisory primary, re-resolve the decision with it requested, fail closed when it is no longer eligible, record selected_by, and keep the advisory portfolio on the decision.

Change

turn.py calls the shared owner instead of repeating the rule:

decision = apply_controller_advisory_primary(build_turn_decision)

No behavior change. turn.py is back to 1111 lines, under its 1114 baseline, and the fail-closed rule now has exactly one home so the two Turn subcommands cannot drift.

Evidence

• python examples/cli-command-module-size-ownership-command-modularization-smoke.py → ok (this is the red check).
• pytest tests/test_loopx_turn_driver.py tests/test_loopx_turn_managed_step.py tests/test_turn_default_host_binding.py -q → 109 passed. This includes test_turn_cli_binds_advisory_primary_without_hiding_portfolio, which drives the CLI end-to-end and asserts both selected_by == "turn_controller_advisory_primary" and that the advisory portfolio stays on the envelope — i.e. the behavior this refactor moves.
• pytest tests/test_loopx_turn_driver.py -k advisory_primary -q → 1 passed.
• ruff check loopx/cli_commands/turn.py → clean.
• loopx canary premerge --from-git-diff → merge_gate_passed: true, self_merge_allowed: true; 4 catalog canaries including the maintainability ratchet and CLI output-budget smokes passed.

Boundary

loopx/cli_commands/turn.py only, +5/-17. No new module, no new surface, no budget number was raised or lowered.

[huangruiteng]

Approval conclusion (author-owned PR; GitHub blocks formal self-approval).

Exact head reviewed: a1213edce

Verdict: APPROVE. One file, +5/-17, no user-visible surface, no new module, no behavior change.

What I verified

• The red check is real and reproduced: python examples/cli-command-module-size-ownership-command-modularization-smoke.py fails on latest main with turn.py has 1123 lines, above budget 1114. The regression is mine — 3f916ff38 (#4443) took turn.py from 1094 to 1123 against a frozen baseline of 1114 that was already in place at 3f916ff38^.
• The extraction is not invented for the gate. turn_decision.apply_controller_advisory_primary already owns this exact rule and turn.py reimplemented it inline, step for step: resolve the advisory primary, re-resolve with it requested, fail closed when it is no longer eligible, set selected_by, keep the advisory portfolio. turn_decision.py's own docstring says both Turn subcommands must build through it so they cannot drift. This is duplicate knowledge being collapsed, which is the refactor the size gate is asking for.
• Call counts are unchanged: both the removed inline version and the shared function call the builder once, and a second time only when an advisory primary was returned.
• Behavior is pinned by an existing end-to-end test, not by inspection: tests/test_loopx_turn_driver.py::test_turn_cli_binds_advisory_primary_without_hiding_portfolio drives the CLI and asserts both selected_by == "turn_controller_advisory_primary" and that the advisory portfolio survives onto the envelope. It passes.
• pytest tests/test_loopx_turn_driver.py tests/test_loopx_turn_managed_step.py tests/test_turn_default_host_binding.py -q → 109 passed.
• ruff check loopx/cli_commands/turn.py → clean. py_compile → clean.
• The ratchet smoke is green at the new head (1111 ≤ 1114).
• loopx canary premerge --from-git-diff → merge_gate_passed: true, self_merge_allowed: true; 4 catalog canaries passed including the maintainability ratchet and CLI output-budget smokes.
• docs/reference/protocols/turn-envelope-v0.md documents selected_by=turn_controller_advisory_primary; that value is unchanged.

Boundary: loopx/cli_commands/turn.py only. No budget number was raised or lowered, so the ratchet keeps its existing meaning. No frontend, Lark, or CLI entry point changes, so there is no user-visible surface to read back.

Residual risk: none identified for this diff. full-public-smokes shard 0 on main has two further, unrelated failures that pre-date this branch — cli-help-manpage-smoke (unclassified: ['goal-actions']) and canary-promotion-readiness-smoke (nested install-local-smoke.py exits 1 in CI, while it passes locally on a clean latest-main worktree). Neither is touched here and neither is caused by this change.