Fix difference between mono/multi with glass

A difference between single wavelength and multiplexed mode was occuring for scene contained within a glass and observed from within this glass (a test case we never used...).

The decimation PDF when multiplexing was only applied to the accumulated attenuation. It also needs to be applied to the accumulated radiance resulting from previous Next Event Estimation, which is also decimated.

I also did a bit of cleaning of unnecessary code (never used) in the dielectric kernel.