check_log.py: improve extracting FPEAuditor stack traces

We used to treat FPEAuditor stack traces in the same way as python / CoreDumpSvc stack traces where we would treat the full log file after the "stack trace" keyword as the relevant error information. While that is true when the job ends after printing the trace, like in case of exceptions / segfaults, it is not the case for FPEs. FPE may happen early in the job and the trace is printed, but the job continues uninterrupted.

Fortunately, FPEAuditor stack traces include keywords making them easy to identify and figure out which lines are part of the trace and which aren't. Make use of this information and treat the FPEAuditor traces specially in check_log.py to only extract the relevant lines rather than the full log file.

cc @okumura