Benchmark improvements

[lemmih]

Issue summary

• Peak memory usage. Can be measured with /usr/bin/time -v.
• Validation time in a recent snapshot. Lotus doesn't have this functionality but we can fairly easily add it.
• Time to migrate network versions.
• Time to export a snapshot. Lotus requires more than 64GiB of memory to export a mainnet snapshot. Find a way to set a memory limit. If that's too difficult then we can just measure calibnet export.
• Time to compute the state of a single tipset (using lotus state compute-state).
• Use lotus-shed to benchmark individual functions without running the lotus-daemon.
• Use published Docker images instead of building Lotus on our own Benchmark: Don't manually build Lotus #143
• Create common error handler.
• Check preconditions before running script (enough disk space and ulimit -n is correctly set).
• Delete cloned git repos at the end of script instead of after each client run. (not applicable if/when the task to use published Docker images above is completed)
• Investigate and fix issue with missing gems when using bundle install.
• Other misc improvements/optimizations (e.g., long form command line options, modify private module designations, replace erroneous references to heights with epochs or tipsets).
• Ensure snapshot is deleted (only if downloaded) after a successful run (we only seem to be deleting a downloaded snapshot in case of an error).

Other information and links

Migrated tasks from analogous issue in Forest repo.

[jdjaustin]

Currently working on the task to create a common error handler and simultaneously improving error handling (currently we suppress detailed errors).

[LesnyRumcajs]

The CPU usage should also be added to the result of each action. (average, peak)

[elmattic]

We should be able to import snapshots in their compressed form for both clients now (this will save us some space on the droplet).