Reduce size of nsight report jobs

[charleskawczynski]

Some of the nsight reports in the gpu target pipeline are so massive that my compute is running out of application memory to open them-- I often have to close out of every other application that I have open in order to open them (especially the prognostic edmf one).

It would be nice if we could drastically reduce the size of these reports.

I think that one of the biggest wins is if we could reduce what is recorded during the start. Of the ~840 seconds, the first 540 seconds are all compilation.