taxon/tab embeds: support fast & accurate updates

[synrg]

The idea is to no longer tear apart these embeds with regexes on a freshly fetched copy of the embed from the channel each time they are updated, but rather to:

• Parse the Discord embed only at the very beginning of interacting with the embed into an INatEmbed object.
  • This is a recent work that started in 218a899 following commits.
  • The result will be cached for a while so that several people adding their counts in close succession will see reliable and fast updates.
• Provide abstractions for each iNat object component of the embed (taxon, places & users listed on the embed, etc.) and the embed's current state (e.g. taxonomy shown/hidden).
  • As of writing this issue, only the taxonomy shown/hidden abstraction remains to be coded
• Provide a bare-bones version of the embed that can be efficiently cached (i.e. only a dict representing the embed needs to be cached; the actual taxon/place/user records themselves are cached separately).
  • See https://github.com/synrg/dronefly/blob/eded9e74ff8da6a4bd0539664644c6e86b0269eb/inatcog/inat_embeds.py#L95
• Provide a means to reconstruct the whole embed efficiently from the cached object.
  • The "efficiently" part is somewhat complicated if the embed has aged long enough so that users have updated relevant records in iNat since the embed was first created.
    • The current method preserves the original counts associated with the listed places / users and only updates the totals, as to do otherwise would require re-fetching records that were already recently fetched.
    • To do the same thing now, we will also need to parse & include those counts in the dict, which we don't do yet.
  • This is expected to solve image: taxonomy inserted in wrong spot when counts present #139 since the whole embed will be recreated from scratch instead of relying on finding cut and insertion points in the description to remove / re-add updated material.
• Rewrite the current update code to:
  • Lock & update only the cached, already parsed copy of the embed, avoiding the round-trip out to Discord which is ultimately the cause of the race condition bug in taxon reactions: Simultaneous reactions add both user & place #101 .
  • As before, with the old regex/insertion method, we will only want to do API calls for new material added (i.e. the new user & updated totals) or removed (recalculate totals without the removed user).

[synrg]

This is going to take a while, and some iNat API endpoints are known to return wrong results when given user_id=login1,login2,... as a parameter, throwing the total count of observations off, so we need a mid-term workaround for that instead of continuing to aggravate users with obviously wrong totals:

1108972 still uses the old approach but takes the list of user ids for the total from the user_id=# parameters in each listed URL instead of from each listed login in the table.