test(model-switch): assert fetch_api_models kwargs match current signature (#15243)

[briandevans]

What does this PR do?

Fixes `#15243`.

`_model_flow_named_custom` was extended in `647900e8` ("fix(cli): support model validation for anthropic_messages and cloudflare-protected endpoints") to pass `api_mode` through to `fetch_api_models` so Anthropic-messages and Cloudflare-protected endpoints validate correctly. The test that pins "we must still probe the endpoint when a model is already saved" was never updated for that new kwarg and has been failing on every PR's `test` CI lane since — it's one of the four standing baseline failures I've been calling out in CI notes on every open PR (#15158, #15162, #15173, #15185 at minimum, plus any PR a reporter opens in this window).

Fix

One line, plus a comment pointing future readers at the history so no one strips the kwarg again.

Before:
```python
mock_fetch.assert_called_once_with("sk-test", "https://vllm.example.com/v1\", timeout=8.0)
```

After:
```python

fetch_api_models MUST be called even though model was saved.

api_mode was added to the production signature by 647900e

("fix(cli): support model validation for anthropic_messages and

cloudflare-protected endpoints") — the test fixture has no

api_mode entry so api_mode or None resolves to None.

mock_fetch.assert_called_once_with(
"sk-test", "https://vllm.example.com/v1\", timeout=8.0, api_mode=None,
)
```

The production call at `hermes_cli/main.py:2920-2923` is:

```python
models = fetch_api_models(
api_key, base_url, timeout=8.0,
api_mode=api_mode or None,
)
```

where `api_mode` comes from `provider_info.get("api_mode", "")`. The test fixture has no `api_mode` entry so `api_mode or None` resolves to `None` — the expected assertion kwarg.

Related Issue

Fixes #15243

Type of Change

• 🔧 Test fix (no production behavior change)

Test plan

• All 6 tests in `TestCustomProviderModelSwitch` pass locally
• This failure has appeared on every open PR's `test` CI lane since 647900e landed — the fix clears it from the baseline failure set
• No production code touched

[Copilot]

Pull request overview

Updates a CLI regression test to match the current _model_flow_named_custom() → fetch_api_models() call signature, fixing a long-standing CI failure after api_mode was added to the production call path.

Changes:

• Update the mocked fetch_api_models assertion to include the api_mode=None kwarg (matching production behavior when provider_info has no api_mode).
• Add an inline comment documenting why api_mode is expected to be None in this fixture.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[briandevans]

Closing — this commit already landed on `main` via 5fdd9f6d3 (cherry-picked with authorship preserved). Verified locally: `tests/hermes_cli/test_custom_provider_model_switch.py` passes 11/11 against current `origin/main` (`810d98e89`) — the `api_mode=None` assertion is in place at line 59 and the standing baseline failure on the `test` CI lane is resolved.

No follow-up needed. Thanks!