Expand AbstractDynodeRunner.save_inference_timelines()?

[kokbent]

Currently the AbstractDynodeRunner.save_inference_timelines() is designed to randomly select a few posterior particles, run them, and extract the "timeline" (predicted hospitalization, strain proportion etc.) It was created with fitting in mind, where you get thousands of posterior samples and you just want to run a subset as representative of the overall fit.

However, this function turns out to be useful for just this context. Currrently, it's been used for example, to also run projections of a number of particles. Basically, if you give a set parameters, this function can produce the simulation trajectory.

Some thoughts about expanding (or refactoring?) this function:

1. Make random selection outside of the function: Since the random selection is built into the function, this means I cannot run specific particle within the dictionary. Say all I care is the output of 0_3, then this function is not going to be useful. Sure I can do a hack by making a dictionary of 1 x 1 list but it can be clunky. Ideally, we should let user choose their particle to run, e.g., particle_to_run = [1, 4, 8, 13, 42]. And if they wanted it random, they can write a random sample to do so. (Although this does make it not backward compatible...)
2. Potential to support multiprocessing: we should be able to leverage the multiple cores, in the case of running say tens or hundreds of particles. It may potentially fail in Azure... but should work well locally 🤔