Initial condition support for torchaudio.functional.lfilter

🚀 The feature

scipy.signal.lfilter supports an initial condition zi. This is critical for dealing with streaming inputs where we get one sample at a time. This feature is currently missing in torchaudio.functional.lfilter.

Motivation, pitch

Without support for initial condition, I had to maintain a large sliding window of signals and apply the filter on this entire window every time I get a new input. If the initial condition is supported, then it becomes much more efficient. For my particular use case, I've been porting scipy based code to pytorch and ran into this issue.

Alternatives

No response

Additional context

No response