caches: use context based cache for repository. - due to gevent sharing connections problem we have to make the cache key per each green thread.

cache: fix overwrite of flock timeout, and improve logging.

caches: make gevent curl connection cache friendly. - gevent threads needs it's own cache, and connection - added strict detection of cache problem in gevent curl

caches: use repo.lru based Dict cache. This LRUDict uses Timing Algo to not have to use locking for the LRU implementation, this it's safer to use for dogpile. We used it before with beaker, so it's generally more stable.

http-client: improve exceptions

depdendencies: bumped scandir to 1.8.0

depdendencies: bumped pycurl to 7.43.0.2

caches: reduce timeouts on flock, and provide error output about filename which we cannot lock at.

vcs: add nicer traceback information for curl errors.

caches: use .refresh() instead of .invalidate() - .invalidate() does a delete of cached valued under a key, which means in previous logic we delete, get_or_create, and compute. - .refresh() is simply replacing exising key with newly computed values, this creates less locks, and it's faster.