Improve processing performance

Processing a trace currently takes a long time (tens of minutes and more; hours even). There's currently two different steps for "processing" a trace. Each one takes a long time.

1. Convert trace data to pickle file. This only needs to be done once.
2. Read pickle file and process events one by one.

We should start by profiling the 2nd part for a typical use-case and see what's taking so long and what could be improved.

Also, perhaps the upgrade to babeltrace 2 (#30) can help (e.g. to process traces directly).

#20 (closed) can be a possible solution or a step in the right direction.