Pass bias and dropout args to Conformer convolution block (#2215)

Summary:
Modifies `ConformerLayer` to pass `bias=True` (to be consistent with feedforward network defaults) and `dropout=dropout` (omission was a bug) to the convolution block.

Pull Request resolved: https://github.com/pytorch/audio/pull/2215

Reviewed By: carolineechen, nateanl

Differential Revision: D34164345

Pulled By: hwangjeff

fbshipit-source-id: 59fc804a1fe3b96e69e9fa5a2f9de94194d7bc55

torchaudio/models/conformer.py

@@ -144,6 +144,8 @@ class ConformerLayer(torch.nn.Module):
             input_dim=input_dim,
             num_channels=input_dim,
             depthwise_kernel_size=depthwise_conv_kernel_size,
+            bias=True,
+            dropout=dropout,
         )
 
         self.ffn2 = _FeedForwardModule(input_dim, ffn_dim, dropout=dropout)