Two core client tests are ignored, for different reasons

[charliewwdev]

Both in core/tests/suite/client.rs, both marked #[ignore] with the reason inline.

chatgpt_auth_sends_correct_request covers the hosted-account transport: a bearer token from a login, an account-id header, a request to /api/opencli/responses. This build configures no such provider, so nothing sets those headers. The code it covers is still there for anyone who configures one, which is why the test was kept rather than deleted. It needs a fixture that sets up that provider.

context_window_error_sets_total_tokens_to_model_window fails on the harness, not the behaviour. The mock provider qualifies as OpenAI's own, so the run takes the remote-compaction path and asks the mock for /v1/responses/compact, which it does not serve. Either the mock should serve that endpoint or the fixture should use a provider that does not qualify.

Several sibling tests in the same file were fixed rather than ignored: they had been broken by the neutralising of model fixtures, which replaced slugs that carried capabilities and context windows with a test-model that carries neither.