fix(ui): classify chat helper models with the shared cloud catalog - #6926

Summary

Six cloud-model checks in src/gaia/ui/_chat_helpers.py now use the existing _ui_cloud_provider, which also reads the shared model catalog. Before, a cloud model from a provider GAIA only learns about at runtime, like opencode.*, counted as local in those places.

Why

The review on #6907 pointed out these call sites. They called cloud_model_provider(model_id) with no metadata, which only recognizes the fireworks and amd prefixes. The module already has _ui_cloud_provider for this, and other callers in the file use it.

The most important one is _title_model_for. It exists so "a local chat's content never leaves the PC just to name it." With a guard model as the session model and a runtime-discovered cloud model as the default, that check missed the cloud default, so a local chat could be sent to it just to generate a title.

Linked issue

Refs #4784, which #6907 fixes in LemonadeClient. This PR is independent of #6907 and doesn't touch lemonade_client.py.

Changes

These call sites now use _ui_cloud_provider(model_id):

• _title_model_for: both the title model and the session model.
• _generate_session_title: the local/cloud switch for title request options.
• _title_when_model_free: whether to wait for a local slot.
• _maybe_update_session_title: whether to title at turn start.
• _empty_answer_backend_problem: whether to check that the model is loaded locally.
• The context-size check before a turn: it already checked the health payload's recipe, and now uses the same helper as the rest.

The PR also removes the cloud_model_provider imports those functions no longer need.

Test plan

Run in a python:3.12-slim container with .[ui] and the test tools installed:

• pytest tests/unit/chat/ui/test_chat_helpers.py tests/unit/chat/ui/test_chat_title_on_send.py tests/unit/chat/ui/test_model_slot_discipline.py: 231 passed.
• python util/lint.py --all: passes, with the existing mypy warning count.
• pytest tests/unit: 34 failures, but 17 test workers crashed under load. The 20 test files with failures that main didn't have pass when rerun on their own (1,967 passed). Most of the other failures also fail on main in my environment.

With the source change reverted, both new tests fail:

FAILED test_chat_helpers.py::TestEmptyAnswerBackendProblem::test_a_discovered_cloud_model_skips_the_loaded_check
FAILED test_chat_title_on_send.py::test_a_local_guard_chat_is_not_titled_by_a_discovered_cloud_default
2 failed

Evidence

The two new tests cover the title-model privacy check and the empty-answer diagnosis, with a model known only from the catalog (opencode.big-model). Without the fix, the guard chat gets titled by that cloud model, and the diagnosis checks whether it's loaded locally. With the fix, both treat it as cloud. The other changed call sites use the same helper, and their existing tests pass.

Not tested against a real cloud provider or a running Agent UI.

• Agent exposed in the Agent UI: N/A, no visible change apart from the title and diagnosis behavior above.
• MCP tools / servers: N/A
• CLI: N/A
• HTTP API / REST: N/A

Checklist

• I have linked a GitHub issue above (Closes #N / Fixes #N / Refs #N).
• I have described why this change is being made, not just what changed.
• I have run linting and tests locally (python util/lint.py --all, pytest tests/unit/).
• I have attached real-world evidence matched to the surface I changed (see Evidence above), or marked each surface N/A.
• I have updated documentation if user-visible behavior changed (see CONTRIBUTING.md). N/A: no documented behavior changes.