Torchdatasets

This minor release introduces Python `3.6` compatibility hence the library can now be freely used on [Google's Colab](https://colab.research.google.com/).

Additional changes:

- Improved `related` section in documentation
- Introduced official library alias (`td`, `import torchdata as td`) throughout documentation
- Redesign of `README` & example of `torchvision` integration.

Up to this point only single bug was found (`torchdata.WrapDataset` performing infinite recursion).

This is probably the last `0.1.x` release, next version (`0.2.0`) will focus on improvements of `pipeline` related minor issues, naming and finishing documentation.

This release was focused on missing functionalities, only minor improvements (beside documentation) and coverage this time.

This is the last pre-release before releasing for wider audience.

Major Features and Improvements:

- Refactor concrete `datasets` into `torchdata.datasets` module (only `torchdata.Dataset` and `torchdata.Iterable` are now available inside `torchdata` main namespace)
- `torchdata.Iterable` got `apply` (just like `torchdata.Dataset`) and `filter` abilities
- Add `torchdata.datasets.WrapDataset` and `torchdata.datasets.WrapIterable`, which act as a proxy between existing `torch.utils.data.Dataset` / `torch.utils.data.IterableDataset` and `torchdata` counterparts. Using those classes, one can easily transform `torchvision` datasets or any other pre-made and use `map`, `cache` and a-like on them.
- `torchdata.Dataset` got `reduce` operation

Bug fixes:

- Minor test coverage improvements in uncertain places

Another pre-release focused mainly on bug fixing and maintenance of current functionality. All changes are breaking at this point and API should be considered unstable.

Major Features and Improvements:

- Added `torchdata.modifiers.Lambda` (arbitrary function can modify any `cacher`)
- Made `torchdata.cacher.Pickle` work as a context manager (cleaning cache directory after block)
- Function arguments come first unless `varargs` have to be specified. Mainly changes `torchdata.modifiers` module
- `torchdata.maps.Drop` and `torchdata.maps.Select` return single element instead of single element `tuple`

Fixes

- Added `modifiers` to `__init__`
- Fix incorrect `OverSampler` and `UnderSampler` behaviour
- Fix `torchdata.cacher.Pickle` to use correct binary format
- Minor fixes dictated by unit tests (see appropriate commits)

Hello :smile: ,

This is initial release of `torchdata` library, which currently should be considered as __alpha__.

To see what it's all about, check [README.md](https://github.com/szymonmaszke/torchdata/blob/master/README.md).

To get in-depth info, check [documentation](https://szymonmaszke.github.io/torchdata/).

__Hope you will have more fun with data and PyTorch from now on, cheers__ :100: