Don't convert mask to float32 until we loop over data chunks

[astrofrog]

Now that we add the data to the shader in chunks, we can do the conversion to float32 there