- Catch and handle all errors once per request.
- Remove the `rescuing` blocks from the store proxies; rescuing per-method (read, write, increment) is bad because (a) it may result in undefined behavior, and (b) it will trigger repeated connection timeouts if your cache is down, e.g. N * M * timeout latency where N is the number of Rack::Attack metrics and M is the cache requests per metric.
- Add `Rack::Attack.ignored_errors` config. This defaults to Dalli::DalliError and Redis::BaseError.
- Add `Rack::Attack.failure_cooldown` config. This temporarily disables Rack::Attack after an error occurs (including ignored errors), to prevent cache connection latency. The default is 60 seconds.
- Add `Rack::Attack.error_handler` which takes a Proc for custom error handling. It's probably not needed but there may be esoteric use cases for it. You can also use the shortcut symbols :block, :throttle, and :allow to respond to errors using those.
- Add `Rack::Attack.calling?` method which uses Thread.current (or RequestStore, if available) to indicate that Rack::Attack code is executing. The reason for this is to add custom error handlers in the Rails Cache, i.e. "raise the error if it occurred while Rack::Attack was executing, so that Rack::Attack and handle it." Refer to readme.
- Add "Fault Tolerance & Error Handling" section to Readme which includes all of the above.

- Improvements to readme