Assert that a served call's pick and its pairing are named when they differ

[maverox]

Follow-up to #258 and #260, now on main.

Since a served call is paired with the recording it names (#260), the candidate's pick and the scorer's twin can differ on one path only. An exact hit already owns the named event, so the served call falls back to the search by shape and pairs with another. #258 names that case in three ways:

• the ArgsServedPairingDisagrees kind;
• the served event on the ledger row;
• a warning giving both sequences.

After the stack merged, a review measured the check as completely unasserted. Neutralising it (.is_some_and(|served| served != twin_seq) → |_| false) left the orchestrator suite green, 722 passed and 0 failed, so it could have stopped firing unnoticed.

a_served_claim_on_an_event_an_exact_hit_owns_falls_back_to_the_shape_search already builds that path: the served call names 701, the exact hit owns it, and the call pairs with 702. It now also asserts the three signals:

• the row on 702 names 701 as the event the call ran on;
• the scorecard counts the kind once;
• the warning names both sequences.

The note on the end-to-end test that recorded the gap now points at this test.

Evidence

• Three mutants, each against the whole workspace with --no-fail-fast, all killed:

  mutant	tests that fell
  the disagreement check neutralised, as the review did	this test
  the ledger row's served event dropped	this test, and the end-to-end test
  the warning dropped	this test

• just verify: exit 0, clippy -D warnings clean, the whole suite passes, web 79/79.

• Tests only; no production code changes.