Delete the byte-identical second declaration of srv3_wiring_with_a_duplicated_header, keeping both authored rationales

[gunbai-bot]

MAIN IS RED. claim_executor --required-lane witnesses refuses the floor at
dag/gunbc/fleet_fan_wiring_witness.dag:754:1: error: duplicate declaration 'srv3_wiring_with_a_duplicated_header' in module 'gunbc.fleet_fan_wiring_witness' -- a second declaration of one name silently replaced the first (run 33096015791,
3bbd53c05a5).

MEASURED, not inferred: the two declarations at 447-453 and 754-760 are
BYTE-IDENTICAL -- diff over the exact seven-line ranges returns empty. So there
is no naming decision and no discrimination decision here. The two consumers, at
236 and 533, receive the same value whichever declaration survives, because there
is only one value. A dispatch describing these blocks as differing was RETRACTED by
its author on the same measurement; the difference reported there was the following
declaration's text, pulled in by a fourteen-line range over a seven-line block.

WHAT DIFFERED WAS THE ANNOTATION, and both halves are irreducible rationale under
DESIGN 4c, so neither is dropped: the survivor now carries the membership-vs-
multiplicity reading from the first and the FAN2 note from the second -- srv3 has
no FAN2 row, so prepending one would have produced an UNEXPECTED header rather than
a duplicated one, which is what a first draft of this control got wrong. That note
is about how to author this fixture and is lost with the block it sat above.

The unobserved-header control attributed to the second block is NOT that
declaration: it is srv1_wiring_with_unobserved_driven_count, the next one, and it
is untouched here.

WHAT THIS DOES NOT FIX, stated because a green is not available from this change
alone: the same floor run reports five further findings in
dag/product/fabric/contention.dag (an absent import member grant_duration_seconds,
two non-exhaustive matches, two unresolved calls), and the declarations phase fails
on the first of them. Those are a separate defect with a separate owner. This PR
removes exactly one of the six findings and main stays red until that one lands.

🤖 Generated with Claude Code

[gunbai-bot]

A red check on this PR is expected and is not this diff.

A `pull_request` run builds the MERGE REF, so main's redness enters this run whatever the diff contains. Main is red on six floor findings; this PR removes exactly one of them. The other five, and the entire `declarations`-phase failure, are dag/product/fabric/contention.dag and are owned by #9488, which is open and mergeable. Green is not available to this branch until that lands.

One correction to how I stated the population, worth recording because it applies to any list this floor prints: the six is a PREFIX, not a population. #9488 also repairs contention.dag at 456/466, where offer_quoted_total_for_grant's result is consumed raw as a MoneyAmountMicro after #9397 made it a QuotedTotal coproduct -- a seventh site the run never printed. The same truncation is visible one level up, where the declarations phase printed one finding against the floor's six, because it stops at the first refusal. Anyone repairing exactly the printed six would have gone red again on a site that reads as a fresh defect rather than the remainder of this one.