Add persistent per request degradation on failure

[bourivouh]

When bad things with redis happens, it raises a lot of Timeout or Connection error. If DEGRADE_ON_FAILURE is enabled, django goes to database after every error and can get its data anyway. But for every db request it does three (and even four if redisreplica is used) queries: two for asking replica and master for a cached data, one for getting data from db, and another one for saving cache.
In this fix client will be marked as degraded if Timeout or Connection error happens. After that no queries will be executed by this client during current request. So in case which described above we will have only three queries per first fetch (asking replica and master for cached data and getting data from db), and we won't try to ask cacheops for all next queries untill request won't be finished.

It's disabled by default and can be enabled by setting CACHEOPS_DEGRADE_PERSISTENT_PER_REQUEST flag to True.

[bourivouh]

@ttyS15, @tumb1er I think it can be interesting for you

[tumb1er]

CACHEOPS_DEGRADE_WHOLE_REQUEST?
CACHEOPS_DEGRADE_TILL_REQUEST_FINISHED?
I read these settings like "Cacheops, honey, please, degrade on failure!" or "Cacheops, please, degrade till the end of request" :)
Or like a term:
"persistent degrade for whole request".
I mean, proper english grammar is necessary here.

[bourivouh]

CACHEOPS_DEGRADE_TILL_REQUEST_FINISHED sounds ok.
Also there is an another bug here: I shouldn't change default for CACHEOPS_DEGRADE_ON_FAILURE