Increase the precision of datasets

[mzur]

Quimbi reads a dataset as many PNGs, where one of the four RGBA channels of a PNG contains the values for one mass channel. This means that, for each PNG, four mass channels can be stored. However, one channel of a PNG is limited to 8 bit values. So the floats (or doubles?) of the original dataset are squeezed into 8 bit values, loosing precision. The 256 different values of 8 bit numbers are not enough to resolve differences between mass channel intensities that are of interest for the biologists.

Find a way to increase the precision to 16 or 32 bit in Quimbi. One mass channel could be stored in two (four) channels of a PNG to store 16 (32) bit values. This would require a new layout and access mechanism in glmvilib. Alternatively it could use RGBA floating point textures or WebGL 2.0 (if the migration is not too hard).