Add config option mongo_count_timeout to skip the global count per request

[ml-evs]

It seems that our mongo implementation is very slow for large collections, in part because of the global structure count required for each filter. This PR adds the ability to disable that (and thus disable data_returned).

cc @eimrek,

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@22f51a1). Click here to learn what that means.
The diff coverage is 77.77%.

@@            Coverage Diff            @@
##             master    #1757   +/-   ##
=========================================
  Coverage          ?   90.77%           
=========================================
  Files             ?       74           
  Lines             ?     4627           
  Branches          ?        0           
=========================================
  Hits              ?     4200           
  Misses            ?      427           
  Partials          ?        0           

Flag	Coverage Δ
project	90.77% <77.77%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
optimade/server/config.py	93.75% <100.00%> (ø)
...made/server/entry_collections/entry_collections.py	95.83% <100.00%> (ø)
optimade/server/routers/utils.py	95.12% <ø> (ø)
optimade/server/entry_collections/mongo.py	91.42% <71.42%> (ø)

[JPBergsma]

Maybe I do not fully understand what you are writing here, but I think MongoDB should know how many entries there are in a collection, so for an empty filter the query should not be slow. For a more complex query for which MongoDB cannot drastically reduce the number of entries using one of the already existing indexes this would still be a useful feature though.

[ml-evs]

You are right, it shouldn't be slow, but previously for an empty filter we were just naively calling count_documents which does still do a full scan that can be very slow. Now, I am using estimated_document_count for this case, which uses simple collection metadata to just return the number. However for filters, it can still make use of this timeout.

[eimrek]

Gave this a try on the MC server for the big database, and everything works well.

See https://dev-optimade.materialscloud.org/archive/li-ion-conductors/v1/structures

Would be good to have this merged. Thanks!

However, the links:next is still broken for our APIs, but i realized it's broken also for releases 0.25.1 and 0.25.2, so it's not related to this change.