Allow RedisCacheStore to fail safely when maximum client connections reached

[bmedenwald]

This fixes #37337

Summary

Recently I encountered a service interruption relating to Heroku Redis, which I use for caching. The Redis server returned ERR max number of clients reached. Redis Cache Store has failsafes to make sure data is still returned when Redis suffers connection errors, but the max number of clients reached error is not a Redis::ConnectionError, it is instead a Redis::CommandError. Instead of being returned fresh data from the database, my users were sent 500 errors.

This fix simply changes the rescue to save requests from all Redis errors. Of course, errors are still reported to the error handler that can be registered so errors will still be sent via the method a user chooses. This is consistent with the other caching stores, which save from virtually all errors:

• MemCacheStore rescues from all Dalli errors (Dalli::DalliError).
• ReadThis (another Redis cache) rescues from Redis::BaseError.
• FileStore rescues doesn't declare a specific type and rescues from all.

[bmedenwald]

@jeremy Any chance we could get this merged in for the next minor release? I could really use this on Heroku given the state of their Redis servers lately.

[bmedenwald]

I've added #37337 to outline this fix with the hope that this can make the 6.0.1 build of Rails.