Updated read_10x_h5:
- Renamed the original `read_10x_h5` as `_read_legacy_10x_h5`;
- Added `_read_v3_10x_h5` to read the new Cell Ranger output format;
- The new `read_10x_h5` determines the version of HDF5 input by the presence of the matrix key, and wraps the above two functions. In addition, it takes a `gex_only` argument which filters out feature barcoding counts from the outcome object when it is True (default). Otherwise, the full matrix will be retained.
- For CR-v3, `feature_types` and `genome` were added into the outcome object as new attributes.

Updated read_10x_mtx:
- Renamed the original `read_10x_mtx` as `_read_legacy_10x_mtx`;
- Added `_read_v3_10x_mtx` to read the new Cell Ranger output format;
- The new `read_10x_mtx` determines the version of matrix input by the presence of the `genes.tsv` file under the input directory, and wraps the above two functions. In addition, it takes a `gex_only` argument which filters out feature barcoding counts from the outcome object when it is `True` (default). Otherwise, the full matrix will be retained.
- For CR-v3, `feature_types` was added into the outcome object as a new attribute.

Added test data and code for the revised functions.

Note for the genome argument:
- There is a genome argument in Scanpy's `read_10x_h5` function but not in `read_10x_mtx` as the genome was already specified by the path of input directory. The outcome object of the two functions should be the same which always take one genome at a time.
- In this PR, when there are multiple genomes (e.g. Barnyard), `read_10x_mtx` always read them all, whereas `read_10x_h5` always need to specify one of them (mm10 by default). However, when `gex_only == False`, the `genome` argument will be ignored and the whole matrix will be read.

Let Scanpy read from Cell Ranger 3.0 outputs (#1)

Merge with master

update to match upstream