Benchmarker improvements

[nathanwbrei]

Some improvements to the benchmarker following the latest tests:

• Set default parser_ms = 0, in order to avoid the scaling curve flattening at ~100Hz
• Inform user AFTER running benchmarker what their options are for tuning the benchmarker (as opposed to in the docs)
• Automatically generate plots from inside the script that loops over the NUMA strategies
• Fit functions to the scaling curve and calculate the parallel fraction as per Amdahl's Law

[nathanwbrei]

Also there's a weird segfault when we run the benchmarker with jana:nevents=1000. Note that there's jtest:nevents, jana:nevents, and a JBenchmarker-set NEVENTS parameter (The latter was clearly intended to override any user-set jana:nevents.)

[nathanwbrei]

Here is the current status:

• Got rid of the redundant jtest:nevents and NEVENTS parameters
• Set default value for jtest:parser_ms to 0. (Same for plotter; I also added a missing critical section)
• Added a log message at the beginning of the benchmarking run informing the user how to tune the benchmarker

Some things I kind of want to do before closing out this issue:

• JTest uses SetParameter instead of GetDefaultParameter, which means that the parameters won't show in the table printed at the beginning of the run
• The segfault is still there, although it has been masked by JBenchmarker
• Still no curve fitting or parallel fraction estimation
• Not sure where the script for generating the NUMA plots lives
• The generated plots have useless titles