Filtering defaults for 10X

Through working with the Tabula Muris data I found:

1) No filtering of the data: does not work for scran: too many samples need to be dropped because of negative size factors.

2) The default filtering used for smart-seq2 is too stringent. For example: ~17000 samples were filtered down to ~8000, and ~27000 genes were filtered down to ~2000 genes.

### Question to answer:
What's a good starting point for 10X filtering?

Start by examining the filtering behavior described for Tabula Muris 10X data and comparing it with the existing smart-seq2 default. Check how the no-filtering case interacts with scran, then document a justified starting point for 10X filtering and the expected effect on samples and genes.

Written by the indexing model from the issue text.