fix(tests): prevent engine loop from calling step() in abort/cancelation tests

[fry69]

This fixes two new errors when running

pytest tests/test_engine_core.py

Please advise if this is the correct approach. For my still limited understanding about the codebase, it sounds plausible.

Below is the message from my coding agent:

---

MockModel lacks .layers attribute, causing scheduler.step() to fail with AttributeError from make_prompt_cache. Adding engine.scheduler.has_requests = lambda: False prevents the engine loop from running step() in tests that test abort/cancellation logic without needing model inference.

Fixes test_abort_request_wakes_blocked_stream_outputs and test_generate_cancel_aborts_request which were failing because the engine loop error handler interfered with expected error messages and collector cleanup.

---

Update: I asked the agent:

Would it be a better approach add the new .layers attribute to the MockModel to fix the root cause of this problem, rather than preventing the symptoms? Or is that a worse approach? Please explain.

It didn't like this approach much:

---

Both approaches have trade-offs. Here's the analysis:

Approach A — Add .layers to MockModel:

This fixes the make_prompt_cache crash at cache.py:34 but doesn't stop there. The prefill pipeline immediately continues to create mx.array(prefill_tokens), call self.model(input_arr) (which returns a MagicMock), then tries to extract logits, embeddings, etc. from that mock object. You'd need to mock ``.layers, .get_input_embeddings, the forward pass return type, and potentially more — it's a deep rabbit hole with fragile, high-maintenance mocks.

Approach B — has_requests = lambda: False (current fix):

These tests test abort/cancellation logic, not token generation. Preventing step() from running is semantically correct — the engine loop shouldn't be trying to generate tokens with a mock model. The pattern is already established at test_engine_core.py:586, so it's consistent.

Approach B is better here because it's simple, targeted, and doesn't require building fragile mock infrastructure that would need ongoing maintenance as the prefill pipeline evolves.