[Bugfix] [MoE] fix error in deepseek when using allgather (#3824)

### What this PR does / why we need it?
After refactoring vllm_ascend/models and FusedMoE, we are unable to pass
`gate` from deepseekv2.py to `AscendFusedMoE.forward`, which will result
in error when running deepseek v3/r1 with allgather.
Hence, this pr removes `gate` related computations from FusedMoE module
in eager/aclgraph mode.
### Does this PR introduce _any_ user-facing change?
`rm_router_logits` is deprecated in eager/aclgraph.
### How was this patch tested?
e2e & ut

- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/releases/v0.11.1

Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>

tests/ut/ops/test_prepare_finalize.py

@@ -199,13 +199,8 @@ class TestPrepareAndFinalize(unittest.TestCase):
         hidden_states = torch.randn(3, 8)
         router_logits = torch.randn(3, 2)
 
-        # Mock the gate function for rm_router_logits=False case
-        mock_gate = MagicMock()
-        mock_gate.return_value = (router_logits.repeat(2, 1), None)
-
-        h_out, r_out, _, context_metadata = layer.prepare(hidden_states,
-                                                          router_logits,
-                                                          gate=mock_gate)
+        h_out, r_out, _, context_metadata = layer.prepare(
+            hidden_states, router_logits)
 
         # After all-gather with DP=2, should double the batch size
         self.assertEqual(h_out.shape[0], 12)
@@ -266,14 +261,8 @@ class TestPrepareAndFinalize(unittest.TestCase):
         hidden_states = torch.randn(3, 8)
         router_logits = torch.randn(3, 2)
 
-        # Mock gate for router logits recomputation
-        mock_gate = MagicMock()
-        mock_gate.return_value = (torch.randn(7, 2), None)
-
         # Run prepare
-        h_out, r_out, _, _ = layer.prepare(hidden_states,
-                                           router_logits,
-                                           gate=mock_gate)
+        h_out, r_out, _, _ = layer.prepare(hidden_states, router_logits)
 
         # Should be global tensor: [7, 8] and [7, 2]
         self.assertEqual(h_out.shape, (7, 8))