§3 de-fork: rename dsl std.verification record TestClaim -> AssertionClaim (dissolves A1 EqualsClaim closure-poison)

[briansrls]

Summary

§3 single-authority de-conflation: rename the record type TestClaim → AssertionClaim in dsl/std/verification.dag. Two genuinely-different concepts were wrongly sharing one unqualified name:

• std.verification (dsl) type TestClaim { kind: AssertKind, label: String } — a test assertion model. → renamed AssertionClaim (the kind field is an AssertKind, so the new name is truthful).
• v2.std.verification (src/v2) type TestClaim = … | EqualsClaim | … — the v2 claim coproduct. Stays canonical, untouched.

This is a rename only — not a unify, and it deliberately does not touch the module std.verification name-fork or any dsl↔v2 std de-fork (separate operator-strategic lane); it stays compatible with whatever that lane later decides.

Why

This was the root cause of the A1 (#5615) type 'EqualsClaim' not found in scope floor failure (reverted by #5627). The resolver's per-module scope (env.bindings) is keyed by unqualified type name. When a PR expands a test's transitive import-closure to include both verification modules, the TestClaim slot silently resolves to one binding — the dsl record won, and since variant-constructor registration only fires for coproducts (Disj), EqualsClaim/StructuralEqualsClaim were never registered → "not found in scope". A1 was an innocent trigger: it dragged std.verification(dsl) into the floor test's closure via gunbc.ci_failure_class → extdeps.cache.sccache → std.cache_interface → std.verification. The unqualified name-clash is the landmine; this rename removes it so the name denotes one concept tree-wide.

(The deeper §5 resolver fail-open — silently picking one binding instead of failing closed on a type-name collision — is operator-gated and out of scope here.)

The dsl record had zero external consumers of the TestClaim/TestCase names (the 5 import std.verification sites all import only the NanosecondDuration/nanosecond_duration* cost helpers), so the change is the 2 lines in dsl/std/verification.dag (the decl + the TestCase.claims field).

Test plan — proven by execution (instrumented oracle + both source-root orders)

On the merge-result tree = current-main + this PR + A1 (re-applied locally for the test only), via claim_batch resolving src/v2/workflow/affected_set_floor_runner_test.dag:

• Oracle (DBG_VARIANT): TestClaim binding connective=Disj, children [CompilesClaim, DiagnosticClaim, EqualsClaim, StructuralEqualsClaim, RoundTripClaim], EqualsClaim_in_locals=true, collisions=0 → PASS (before this PR: Conj{kind,label} / EqualsClaim_in_locals=false → FAIL).
• dsl-first (--source-root dsl --source-root src/v2): PASS (56 modules, 1601 resolved items).
• v2-first (--source-root src/v2 --source-root dsl): PASS (56 modules, 1601 resolved items).

Collision dissolved in both orders; the closure still includes std.verification(dsl) but the renamed AssertionClaim no longer shadows the v2 coproduct.

Diagnosis context: dashboard node adhoc-f83d03b9-69b; parent-signed the std rename (read-only constraint lifted for this edit only; resolver remains off-limits).