Maybe Andrew knows how to detect this -- if it is just a matter of looking at each plugin to see if it looks like the word "float" is in effect for the majority of variables, that is something I could do. Or if Andrew has another way?
may be look for lines like
case BC_RGB_FLOAT
case BC_RGBA_FLOAT
with grep -r in plugins directory?
it looks like they all support at least input in float colorspace but histogram for example calculates curves with 8/16 bit precision, may be this is that gives clamping at the end? Does it exist if you output to native tiff?
also, you can look for
grep "CLAMP" -r plugins/
clamp macro itself, but you need some code reading for understand what exactly it clamps ... for example all display output still 8bit, so there is no sense yet to keep full range of channel values for in-plugin drawing ...
sorry for absence of easy answer ...