Reduce duplication when using multiple Notifiers

[Swatinem]

As you can see in this trace
each notifier instance is doing a bunch of duplicated work:

• It is re-requesting a github token over and over again, adding to our rate limiting problem
• It is querying (presumably the same) Commit over and over again
• The SELECT/INSERT of the commit_notifications could most likely be optimized by batching across all the various status notifications to be emitted
• POST-ing the notifications to github should ideally happen in a batch (if the API allows it), instead of doing it one-by-one

Another optimization opportunity:
We have a redis GET immediately followed by SETEX is likely some form of cache. However it has the following problems:

• GET with an immediate SETEX indicates a cache miss, so the cache might be ineffective
• There is no visible gap / span in between the GET and SETEX, so whatever operation is being cached is likely very fast and does not need caching in the first place.

[Swatinem]

Some update here:

• The code was refactored to reuse the same "repo provider" instance across all notifiers.
• That, along with caching improvements seems to have solved the problem of re-requesting the github token over and over.

I have tried to batch the SELECT/INSERT of commit_notifications in codecov/worker#1189, but that is blocked by some concerns around task timeouts/retries.

The GET/SETEX redis calls used for caching are there to avoid posting identical comments/statuses multiple times.