[add:lib] Add sequentia.datasets module

[eonu]

Implements #141 and partially #194 by adding sequentia.datasets.

Major changes

• Add sequentia.datasets module:
  • sequentia.datasets.load_digits for loading the Free Spoken Digit Dataset (stored in lib/sequentia/datasets/data/digits.npz).
  • sequentia.datasets.base.Dataset for representing generic dataset objects.
  • sequentia.datasets.base.TorchDataset for representing torch-compatible datasets.
  • Add RTD documentation and code examples.
  • Update notebooks to use sequentia.datasets.
• Modify _Validator.is_observation_sequences() to accept a specific dtype for numpy.ndarray objects (defaults to numpy.float32).
• Use numpy.float64 for KNNClassifier.
• Move check_package into lib/sequentia/internals/versions.py and also add is_torch_installed() for checking torch versions.
• Remove torchaudio, torchvision and torchfsdd dependencies.

Minor changes

• Remove sentiment classification example from README.md.
• Shorten RTD preprocessing warning about torch compatibilty.
• Add notebooks/nbutils.py with a helper function play_audio for playing audio samples from numpy.ndarray objects.