[AMD] Add moe_ep_size/moe_tp_size to the allreduce-fusion gate test mock

[kangwangamd]

test_aiter_allreduce_fusion_amd.py::TestAiterAllreduceFusionGate has been red on both pr-test-amd and pr-test-amd-rocm720 since Aug-12 (log):

File "python/sglang/srt/layers/communicator.py", line 828, in should_fuse_mlp_allreduce_with_next_layer
    if parallel.moe_ep_size > 1 and parallel.moe_tp_size > 1:
AttributeError: 'types.SimpleNamespace' object has no attribute 'moe_ep_size'

#30700 added a hybrid EP+TP check to should_fuse_mlp_allreduce_with_next_layer, so the gate now reads moe_ep_size and moe_tp_size off get_parallel(). The mock in this test only provides tp_size, so every case in the class raises before reaching its assertion.

This adds both attributes to the mock, set to 1. That is the dense-TP configuration these tests are written for: the new guard only short-circuits when both are greater than 1, so 1/1 leaves test_dense_tp_fuses asserting the fusion path as before, and the other cases still exercise the DP-attention and EP-backend branches they were added for.

Test-only change; no source behaviour is affected.

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

This repairs an existing test rather than adding one, and there is no documentation or performance impact.

---

CI States

Latest PR Test (Base): ❌ Run #32101495165
Latest PR Test (Extra): ❌ Run #32101494926

[kangwangamd]

@wenscarl could you take a look? This is a test-only fix for the moe_ep_size AttributeError that has had test_aiter_allreduce_fusion_amd.py red on both AMD PR workflows since Aug-12.

[kangwangamd]

Closing in favour of #35340, which makes the same change (moe_ep_size=1, moe_tp_size=1 on the get_parallel stub) and is already in @bingxche's hands for CI.

One data point in case it is useful: on stage-c-test-large-8-gpu-amd-rocm720 shard 3 (run 31916881653) the shard reports Test Summary: 3/4 passed — this file is the only failure, and all 4 tests in the class error out before their assertions. So the stub fix should green the whole 8-GPU shard.