limit buffer for mongo backend

[JoranAngevaare]

What is the problem / what does the code in this PR do
Address #346

Can you briefly describe how it works?
Keep track of how many runs (datakinds) are stored in the buffer. If to large, dump some.

MWE

for r in range(10165, 10165+10):
    if st.is_stored(f'{r:06}', 'online_peak_monitor'):
        st.get_array(f'{r:06}', 'online_peak_monitor')
        print(st.storage[-1].backends[0]._buffered_backend_keys)
        print(sys.getsizeof(st.storage[-1].backends[0].chunks_registry)/1e6)

[darrylmasson]

Nesting a for-loop inside a while-loop might be really inefficient, but probably won't be a real issue unless someone decides to change the defaults.

[darrylmasson]

Python's complained at me if I modified a dictionary while iterating over it. List-casting the keys probably helps with this, but I got around it by storing the keys to delete in a second list.

[JoranAngevaare]

you really have a good memory for errors, got to give you that. This is true.

However if you look carefully we are making a copy of the keys by iterating them when converting to a list. Therefore you are not iterating over the keys but over the list of strings that are extracted