Pystore

------
- Added support for `dask.distributed` via `pystore.set_client(...)`
- Added `store.item(...)` for accessing single collection item directly (pull request 44)
- Added `store.set_partition_size(...)` and `store.get_partition_size()`. Default is ~99MB.

------
- Updated PyPi install script (lib is the same as 0.1.16)

------
- Fixed `npartition=None` issues on `.append()`

------
- Fixed append issues
- Raising an error when trying to read invalid item
- Fixed path issued (removed unnecessary os.path.join calls)

------
- Auto-detection and handling of nano-second based data

------
- `collection.reload_items` defaults to `False`
- Default `npartitions` and `chunksize` are better optimized (~99MB/partition)
- `collection.apply()` repartitions the dataframe based on new data size (~99MB/partition)
- Option to specify the default engine for the store by specifying `engine="fastparquet"` or `engine="pyarrow"` (dafaults to `fastparquet`)
- Solving `fastparquet`/`numba` issues when using Dask >= 2.2.0 by importing `numba` in `__init__.py`