Composite UMAP

[jimmymathews]

Add a composite UMAP, with different color color-coding for each channel, that summarizes all of the channels' worth of intensity values.

To implement this, you need:

1. A mapping of channels to colors. Stored somewhere in the database.
2. A place to store the new composite image (currently doesn't fit in the umap_plots table).
3. A way to get (1) and (2) back out via API call.

[jimmymathews]

There is no clear way to merge the intensity values across channels for each cell to provide a single coloration.
A natural way to deal with this is to work with the dichotomized channels, coloring a cell only if it has 1 for the channel, with ties broken by the intensity values.
However, I'm pretty sure that in most cases people will prefer to see the phenotype ("composite phenotype") represented in the composite UMAP, not the individual channel intensities.
To implement this, one needs to do nested iteration, over all cells and over all phenotypes, checking for membership and assigning a color based on some average of assigned colors per-phenotype.

I think we settled on the plan to use UI elements that enable overlay/non-overlay of selected composite phenotypes (not single channels, which are well-represented in the other UMAPs). So the above comments about reaching back into intensity values to break "ties" and decide on a cell's color are not applicable. What will be needed is a special-case way to determine how to color a cell which belongs to multiple phenotypes: For this I suggest random assignment. This case is not intended to be particularly common.

[jimmymathews]

This will be supported by a new API endpoint that provides the dimensionally-reduced position information for a subsample of each study.