Move in 10x readers

[ivirshup]

Draft PR to move in 10x readers from scanpy. The goal here is to make these readers available without needing all of scanpy's dependencies. Previously this had been blocked by scanpy's use of pytables for these readers.

Previous issues

• Why is the 10x h5 reader implemented in scanpy and not in anndata? #195
• gex_only=True defaults in sc.read_10x_h5 scanpy#1949
• Spin off read/write into it's own package? scanpy#1387

@gtca, you've mentioned some issues with differences between the h5 and mtx backends, are those addressed here?

TODO

• Check if this is where these are going
• Filter feature types during read, instead of after they've been read (if possible)
• Figure out semantics for feature_types argument (how to indicate "only", how to indicate "all")
• Either add var_names to read_10x_h5 or remove it from read_10x_mtx
• Flesh out plan for deprecation of scanpy functions

[codecov]

Codecov Report

Merging #883 (694a103) into master (3c5f63d) will increase coverage by 0.38%.
The diff coverage is 92.23%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #883      +/-   ##
==========================================
+ Coverage   83.64%   84.02%   +0.38%     
==========================================
  Files          34       34              
  Lines        4487     4600     +113     
==========================================
+ Hits         3753     3865     +112     
- Misses        734      735       +1     

Impacted Files	Coverage Δ
anndata/__init__.py	100.00% <ø> (ø)
anndata/_io/__init__.py	100.00% <ø> (ø)
anndata/_io/read.py	84.21% <92.23%> (+6.48%)	⬆️
anndata/_io/specs/registry.py	95.23% <0.00%> (+0.55%)	⬆️
anndata/utils.py	84.95% <0.00%> (+0.88%)	⬆️
anndata/_io/utils.py	73.83% <0.00%> (+0.93%)	⬆️