a CLI param to control heuristic determination of approximate X data distribution

[signechambers1]

Add a new CLI parameter which allows the user to force cellxgene to assume a particular approximate X data distribution, for the purposes of calculating differential expression statistics. See also #2211

Parameter name: --X-approx-distribution

cellxgene launch --help
...
Options:
--X-approx-distribution [auto|normal|exponential]

Values:

• Auto - "sniff" the data, applying a heuristic to determine distribution.
• Normal - assume the data is normally distributed (what we commonly call "logged")
• Exponential - assume the data is exponentially distributed (what we commonly call "not logged"), and apply log1p() before calculating statistics.

Example which forces cellxgene to assume the data is not logged (ie, exponentially distributed, and will be logged before computing diffexp):

cellxgene launch --X-approx-distribution=exponential mydata.h5ad

Example which forces cellxgene to assume the data is normally distributed, and will not be logged before calculating differential expression:

cellxgene launch --X-approx-distribution=normal mydata.h5ad

The 'auto' value is default and will apply a heuristic to determine the approximate distribution of X values.