Make Casanovo model compatible with torch.compile

[Lilferrit]

When attempting to do a sequencing run using a compiled Casanovo model I encountered this runtime error:

...
torch._dynamo.exc.TorchRuntimeError: Failed running call_module L__self___encoder_transformer_encoder(*(FakeTensor(..., device='cuda:0', size=(128, 121, 512)),), **{'src_key_padding_mask': FakeTensor(..., device='cuda:0', size=(128, 121), dtype=torch.bool)}):
strided nested tensors are not supported by meta conversion

from user code:
   File "C:\Users\gavin\OneDrive\Documents\casanovo\casanovo\denovo\model.py", line 844, in predict_step
    self.forward(batch[0], batch[1]),
  File "C:\Users\gavin\OneDrive\Documents\casanovo\casanovo\denovo\model.py", line 210, in forward
    return self.beam_search_decode(
  File "C:\Users\gavin\OneDrive\Documents\casanovo\casanovo\denovo\model.py", line 240, in beam_search_decode
    memories, mem_masks = self.encoder(spectra)
  File "D:\anaconda3\envs\casanovo_env\lib\site-packages\torch\nn\modules\module.py", line 1541, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\anaconda3\envs\casanovo_env\lib\site-packages\depthcharge\components\transformers.py", line 105, in forward
    return self.transformer_encoder(peaks, src_key_padding_mask=mask), mask

From the best I can tell this specific error is an issue with PyTorch's built-in TransformerEncoder module, which is known to be incompatible with torch.compile: https://discuss.pytorch.org/t/runtime-error-when-running-inference-on-a-compiled-nn-transformerencoder/198010

In order to run with a compiled Casanovo model I added self.model = torch.compile(self.model) to the end of ModelRunner.initialize_model.

From my understanding making the Casanovo model compatible with torch.compile would be useful for addressing #328.