Include flag in SelexDataset to use sparse counts instead of non-sparse counts (default)

[ilibarra]

• New flag sparse=True.
• Two tests:
  • Conversion of int to sparse does not generate dataframes with different shape
  • The values between a non-sparse and a sparse object are the same e.g. sum.

[ilibarra]

The flag/conversion is now implemented and in development. However, PyTorch data loader doesn't handle sparse matrices with ease, so we cannot consider this case for extensive testing at this point, unfortunately. We would need something like a sparse to torch conversion during the dataloader step.