[ARRDT] look into optimism thrashing

[alloy]

According to @vladar any read from the cache adds entries to optimism. We need to verify this and try to avoid it, perhaps as described here https://github.com/apollographql/apollo-client/issues/9306#issuecomment-1015634246

[vladar]

Yes, I see optimism cache growing on simple clicks on todo items:

[vladar]

One offender is this call where we create a new fragment document object every time (note: fragment: { is always a new object):

https://github.com/microsoft/graphitation/blob/cc843406967da7c964bd5c68e79e1cc3f1b234f5/packages/apollo-react-relay-duct-tape/src/storeObservation/nodeFromCacheFieldPolicy.ts#L42-L52

A simple caching of this document in a WeakMap (using options.query as a key) allowed to lower growth rate to 2 new entries per click vs 4.

Those 2 remaining items are still to be investigate - I don't think they should appear at all.

[vladar]

One thing that popped up during a discussion: graphitation compiler may produce multiple instances of fragment documents (one in queries, and another for fragment hooks) which may end up in even larger number of optimism entries.

We need to investigate if it actually affects optimism size.

[alloy]

One offender is this call where we create a new fragment document object every time (note: fragment: { is always a new object):

https://github.com/microsoft/graphitation/blob/cc843406967da7c964bd5c68e79e1cc3f1b234f5/packages/apollo-react-relay-duct-tape/src/storeObservation/nodeFromCacheFieldPolicy.ts#L42-L52

This one is resolved by #127.