Easier use of document stream in evaluation

[thopkins32]

Currently, the EvaluationFunction protocol makes some assumptions about how you retrieve the acquired data that was just collected. We should explore what it may look like to access data in-process rather than relying on Tiled or some other service to fetch the data.

You could technically do this now by ignoring the uid argument and writing something like:

cache = []
def document_cache_callback(...):
    """Cache the documents of each run"""
    ...

RE.subscribe(document_cache_callback)

def custom_evaluation(uid: str, suggestions: list[dict]) -> list[dict]:
    """Use the cache instead of uid"""
    ...

It's a bit awkward to do so maybe there is something Blop can do to make this simple.