gstlal_inspiral_marginalize_likelihoods_online: rewrite in python

We have observed several problems with the marginalize likelihoods online job, First it is generally very slow, taking about 15 hours to marginalize across all ~900 bins. Second, it is very prone to errors, for example HTTP errors when accessing inspiral jobs via bottle. And there was no attempt at error handling. Any time the program fails to access a single inspiral job, the whole program crashes and has to start over from scratch. This has caused problems in the past with burn in times and is just generally not good for keeping the dist stats pdf up to date.

To fix the problem, I re-wrote the program in python which should be generally more efficient for loops like this. I also added some error handling so that the program can re-try bins that it failed to access previously. The time for one iteration of this loop has decreased from about 75 seconds to 40-50 seconds with this re-write.