v2 lens tests: stop referencing test fns

[gunbai-bot]

The first area of conversions that clear v2 references to test code before the v2 refusal lands (node adhoc-f8d8160b-c82; owner decision: fix outright, no ledger). The conversion rule is the one the v1 cleanup used.

• lens_application/sg_claims_test: deleted lens_application_introspect_advisory_holds and lens_application_synthesis_gap_polynomial_holds. Each only called another test fn that is already enrolled, and that test still asserts the fact.
• lens_idempotency/sg_claims_test.dag: deleted. Its single claim only called the enrolled idempotency_write_effect_claim_holds.
• lens_cost/p9_…: p9_registry_owner_receipt_holds now asserts only the row count. Name uniqueness is still asserted by its own test fn, p9_registry_fn_names_unique.
• lens_cost/bounded_summation_test: the conjunction moved into the plain fn bounded_summation_expr_authority. Both the test fn and the claim_bounded_summation data row call it.
• lens_structural_resolution/binds_to_resolved_test: deleted data binds_to_resolved_claim_passes. Nothing consumed it, and it called a test fn.
• floor_grandfathered_roster: a typed WitnessDeleted removal for each of the three deleted identities.

Not converted, and deliberately so: the census flagged a few calls outside test directories, but they reach interpreter primitives or plain fns that only share a test fn's name, not test fns. Examples are fallback_arm_census_reconciliation_holds in v2.lens.fallback_arm_census and non_fold_residue_*_red_holds.

Evidence, from a freshly built claim_batch, with PASS counts checked against each file's roster: lens_application/sg_claims 1/1, bounded_summation 3/3, p9_… 2/2, binds_to_resolved 1/1, and floor_eval_step_budget_witness_test (which exercises the roster's removal fold) 27/27.

🤖 Generated with Claude Code