Raman response improvements

[jtravs]

There are two things which could improve the Raman response code in terms of performance and accuracy.

At the moment we use a double length time grid to perform the Raman convolution. This is pretty robust, but is rather expensive. It is somewhat required because for light gases (e.g. H2) the Raman response is very long lived (up to ns), far beyond our computational window.

Some possible improvements:

1. Use a smaller factor than 2 to expand the convolution grid (for higher performance) when the response function is short lived. There is no fundamental reason that the response function and field need the same time grid, and often in fast convolution code (see e.g. DSP.jl) the grid is adapted to he response (e.g. for silica, which has a very fast response, this could lead to a huge saving).

2. use a temporal window function to smooth the frequency response for response functions which do not decay (common in gases, e.g. H2, which never decays on our usual grids)