Main repair: hoist the in-body annotation in microvm_controller_app_key_converge to module-item grain

[gunbai-bot]

dag/gunbc/runner/microvm_controller_app_key_converge.dag carries 7 blocking source annotation sits inside a declaration body. Only module-item grain is modeled errors, landed with #11902 earlier today. Any floor closure that reaches the module refuses.

Why nobody saw it: the required gate's floor job binds no adjudicator, so a refused floor concludes success — the defect #11829 repairs. This is one of the main-side blockers gating that PR. Found by smart-boar-749 while repairing a different class (#11940).

The change: the 7-line annotation block inside clear_staging is hoisted above fn clear_staging, beside the annotation already attached there. Prose unchanged, no semantic edit, no other file touched. Module-item grain is the only grain .dag models for annotations, and the rationale is about the function, so the declaration it describes is unchanged.

Evidence: the file now has zero indented // lines (grep -c '^\s\+//' → 0, was 7). The floor lane on this PR reports the parse phase; read its log rather than the check, since the check cannot go red until #11829 lands.

🤖 Generated with Claude Code

[gunbai-bot]

The repair discriminates, and it uncovers what the parse refusal was hiding

Control: four post-#11902 CI runs (35561659566, 35561459469, 35561411822, 35560693414) each report exactly 7 source annotation sits inside a declaration body errors for this file.
This head (run 35563045570, job 106219416135): required-ci: parse OK 6522 file(s) parse-clean, 0 such errors.

That matters more than one file's tidiness: a parse FAIL lends no declaration index, so before this the floor refused at REQUIRED-FLOOR REFUSAL cause=ArmSetConsumerPlanningUnavailable and never prepared a subject at all. With parse clean the floor now plans (subject=3e550207ade97bb5 modules_resolved=1978 modules_excluded=0).

What that exposed — this module still does not resolve

Not caused by this diff, which moves prose only, but no one should read this PR as making the module clean:

name 'classify_host_file_presence' not found in module 'gunbc.runner_host_file_converge'
function 'classify_host_file_presence' not found in scope
effect summary incomplete: call to 'classify_host_file_presence' in gunbc.microvm_controller_app_key_converge

:58 imports it and :651 calls it; git grep on origin/main finds the name in exactly those two lines and nowhere else — #11902 landed a call to a function that does not exist. It was invisible twice over: behind this file's own parse refusal, and behind the required gate's missing adjudicator (#11829). Dispatched as its own lane (adhoc-e93c96e8-305) with instructions to derive the repair from runner_host_file_converge's existing EntryPresence vocabulary rather than minting whatever compiles, and not to collapse an indeterminate probe into Absent.

Also newly visible on this head and unowned: gunbc.host_memory_observation:166,175 cites gunbc.ci_floor_measurement declarations that are absent, with two matching stale PRE_EXISTING_CITATION_DEBT rows in declaration_index.rs.

— sent from bright-eagle-728