main red again: delete the duplicated roster row (#7288 raced a concurrent identical fix) — 28 rostered vs 27 declared

[briansrls]

Summary

main is red again on the same witness, for the opposite reason. #7288 and a concurrent PR
each independently added the same roster row for ct_render_rust_applied_type_qualified_base_test.
Both merged. The roster now carries the row twice, so
compiler_tests_rust_blobs_are_all_rostered reads 27 declared vs 28 rostered and reds again.

This PR deletes the duplicate — the declaration block and its second language_source_scaffold_roster
entry. Net -6 lines, no other change.

	before	after
declared fn ct_	27	27
rostered rows	28	27
unique rostered blob_decls	27	27

Verified green by execution

PASS compiler_tests_rust_blobs_are_all_rostered
PASS runtime_rust_blobs_are_all_rostered
PASS language_source_scaffold_roster_is_fully_dispositioned

./target/release/claim_batch --entry dag/test/claim/language_source_scaffold_index_test.dag --source-root dag --source-root src/v2, binary built 2026-07-26 13:58 UTC (sha256 a5b9eb4a0643…).

The RED half is already on the record: main's push run at afd33913 failed with
compiler_tests_rust_blobs_are_all_rostered … returned Bool(false) (1 of 2491 discovery witnesses),
which is this exact duplicate.

The finding worth more than the fix: a double-bound name did not refuse

Two top-level data declarations with the identical name
(ct_row_ct_render_rust_applied_type_qualified_base_test) sat in one module and the corpus
compiled and evaluated normally — CI reached the witness and got a Bool(false) verdict, so
nothing refused at compile time.

That is what let a merge race corrupt a roster silently instead of failing loudly. The
namespace-resolution design states the rule plainly — "binding-edge owner, unbound/double-bound =
error" and "no ambiguity at any layer" — so a duplicate top-level binding should be a typed,
located refusal, not a last-writer-wins overwrite. Today it is neither caught by construction nor
by any lens; it surfaced only as an arithmetic mismatch in a downstream coverage count, one merge
later.

Not fixed here (this PR is the unblock, and the wall belongs in the resolver, not in a roster
file). Flagged for the namespace lane: a double-bound top-level declaration must refuse. The
same race that produced this duplicate produces silent shadowing anywhere else in the corpus, and
only this roster happened to have a counting witness pointed at it.

Process note

The duplicate arose because two sessions independently diagnosed the same main-red and shipped the
same one-line fix within minutes. Both were correct in isolation; neither could see the other. The
one-merge acceptance window caught it, as designed — but the double-bound refusal above is what
would have caught it at authoring time instead.