Support compressed data

[cdeil]

Whisper files often have many zeros if not the whole archives are filled, so any compression algo can give large savings.

We should support at least reading gzip and bzip2 files using the built-ins in Python:
https://docs.python.org/3/library/archiving.html

Probably the implementation should read and decompress using BytesIO and then any processing should run on that, and at the end of the read function that bytes buffer is garbage collected.

[cdeil]

Support for gzip added in d616adf

If we add others (e.g. bzip2) we should be compatible with the pandas API for compressed files:
https://github.com/pandas-dev/pandas/blob/4bfe3d07b4858144c219b9346329027024102ab6/pandas/_typing.py#L246-L248