Rethink query search/model update dataflow

[stsievert]

Currently, the data flow updates the model and searches/post queries for the current model:

Here "M_i" means "model i" and "search" clears all queries then searches and posts queries. It basically runs this loop:

future = background(update_model)
for k in count(start=12):
    queries, score = search_queries(num=2**k)
    post(queries, scores)
    if future.done():
        break

[stsievert]

I think a better solution would be to do a complete search:

In this, "score" only scores queries. Posting of queries is hidden from this setup. This implementation would basically require this loop:

queries, scores = [], []
while True:
    f_post = post(queries, scores)
    f_model = background(update_model)
    f_search = background(search_queries, num=2**k, stop=f_model.done)
    queries, scores = f_search.result()
    # ... (updating model, etc)

There's no reason post(queries, scores) can't happen in concurrently with f_model and f_search.