updated rapidpe-rift-pipe to 0.5.1

This fixed two things:

• The previous version works only if the preferred event for a superevent is gstlal. v0.5.0 can choose the highest snr gstlal trigger from the list of events or the gstlal preferred event associated with a superevent if the preferred event is not gstlal.
• In the previous version rapidpe initially look for approximately 500s of data around event time using gwdatafind to write a data cache file (but only a few seconds of data is used for pe). Due to this, in the production run, rapidpe fails to submit a run because 500s of data are not available by the time rapidpe is launched. So, this update reduces the buffer time to 200s of data before the event time and 14s of data after the event time.
• The previous version submitted Rapid PE after about 5 minutes and 45 seconds after the event time. We have reduced this to 30 seconds because that is the maximum amount of time gstlal needs (in ~99% of cases) before their highest snr trigger has settled.
• We have also added the option to run on lower latency data on /dev/shm/kafka/ if it is available, if not it will default to the data it used before which is /ifocache/llcache/kafka/

SCCB: https://git.ligo.org/computing/sccb/-/issues/1231

• update to rapidpe-rift-pipe 0.5.1