Fix TypeError: numpy.float64 cannot be interpreted as integer

[deekshaNVIDIA]

Summary

• Fixes TypeError: 'numpy.float64' object cannot be interpreted as an integer #713: Cast encoder.n_feats_out and similar values to int() across all model constructors and recipe files to prevent TypeError: 'numpy.float64' object cannot be interpreted as an integer.
• Newer NumPy versions return numpy.float64 scalars from arithmetic operations, which PyTorch's nn.Conv1d, nn.Parameter, torch.ones(), and other APIs reject where a native Python int is expected.
• This is a minimal, targeted fix that wraps the filterbank output channel count with int() at the point of use, ensuring compatibility without changing any behavior.

Changes

Core model files (asteroid/models/):

• dprnn_tasnet.py: n_feats = int(encoder.n_feats_out)
• conv_tasnet.py: n_feats = int(encoder.n_feats_out)
• dptnet.py: n_feats = int(encoder.n_feats_out)
• lstm_tasnet.py: n_feats = int(encoder.n_feats_out)
• sudormrf.py: n_feats = int(enc.n_feats_out) (both SuDORMRFNet and SuDORMRFImprovedNet)
• demask.py: n_feats_out = int(encoder.n_feats_out) before passing to helper methods
• dccrnet.py: int(stft_n_filters) // 2
• x_umx.py: int(in_chan) // 2 + 1

Mask network (asteroid/masknn/):

• recurrent.py: int(np.prod(last_encoder_out_shape)) in DCCRMaskNet

Recipe/example files (egs/):

• dns_challenge_INTERSPEECH2020/baseline/model.py: output_size = int(stft.n_feats_out)
• wsj0-mix/DeepClustering/model.py: int(enc.n_feats_out) // 2

Test plan

• Verified the exact reproduction case from TypeError: 'numpy.float64' object cannot be interpreted as an integer #713 (DPRNNTasNet(n_src=2) construction) no longer raises TypeError
• All changes are safe no-ops when n_feats_out is already a Python int
• Checked all usages of n_feats_out across the codebase for completeness

[mpariente]

Thank you !