Feature: Cache for runtime workers

[mtweiden]

This PR adds the ability to interact with a "runtime cache" for workers. Objects that are not CPickle-able (such as Pytorch modules) can instead be loaded to cache. Doing so ensures that objects are only loaded by each worker a single time. The cache is implemented as a simple Python dictionary.

Use:
In a worker process, items in worker cache can be accessed using the following general outline:

from bqskit.runtime import get_runtime

worker_cache = get_runtime().get_cache()

if 'object_name' not in worker_cache:
    object = load_object()
    worker_cache['object_name'] = object
else:
    object = worker_cache['object_name']

[edyounis]

For the most part LGTM, thanks for adding this feature. Please just update the get_cache documentation.

[edyounis]

What is a "worker's local cache"? The documentation here should explain how this can be used while abstracting details of how the runtime works. Consider a user reading this may not know (nor should they know) what a worker is. What is important is that this cache is transient. You can place things in here, to reduce reloading, but they may not be in here later. This informs the user that when using this cache, you should write something like:

def load_expensive_io_model_or_something_or_another(file):
    cache = get_runtime().get_cache()
    if some_file_name_key in cache:
        return that
    # load it; store in cache; return it
    ...

Not saying you can't or shouldn't put implementation details later in the docstring, but the first points should inform the user how to use this method/function/class/etc. In this case, definitely include somewhere that this cache is local to a specific worker and is never serialized/pickled.

[mtweiden]

Removed reference to workers here. Added example as to how to use the cache.

[edyounis]

It's totally me being pedantic, feel free to ignore this completely, but I did want to mention that you do not need to say it's implemented as a dictionary when the type information is right there.

[mtweiden]

Moved references to worker specific cache here.