[jm-rivera]

A no_data part now becomes a gap of its own reason's gap_kind, whatever the answer's diagnostics.timed_out says. The old rule turned every no_data part of a timed-out answer into a timed_out gap. That made sense when Search reported an unfinished part as variable_not_resolved. Search now marks each part it left unresolved at its time limit with the reason timed_out (dfo-apis src/search/engine/fanout.py, _placeholder_no_data), so the override only erased the reason, message and choices of parts Search did answer. The SDK e2e testers saw an entity_ambiguous choice reduced to a timed_out gap with no options.

python/client-contract/README.md states the new rule. answer-timed-out is re-derived so its unfinished part carries reason timed_out and the message dfo-apis renders for it, and the expected gap stays timed_out. The new synthetic answer-timed-out-keeps-reasons keeps variable_not_resolved under diagnostics.timed_out: true and expects a not_understood gap. tools/validate_client_contract.py answer_gaps reads the part's reason alone, and tools/capture_client_contract.py derives both scenarios.

The test row diagnostics-not-an-object-read relied on the old rule, so it is replaced by test_diagnostics_that_is_not_an_object_holds_no_warnings, which pins the README clause through the replay reader with a string, a list and an object control.

Downstream, dfo-apis pins this package and the SDK's contract/SOURCE must sit at or behind that pin. A follow-up bumps both, and the SDK drops its timed-out override in the same change.

Testing: uv run tools/validate_client_contract.py reports 121 scenarios valid; pytest tools/tests 530 passed; python/client-contract pytest 78 passed, ruff check and format clean.