New feature: MongoDB or other NOSQL for memory

The "memory" in TPOT caches candidate pipeline metadata on disk.
The data is cached as a large number of very small ASCII text files each containing data represented by a JSON object.

It occurs to me that this is a crude "document database", and both time and space efficiencies could be gained by using something like MongoDB instead.

I have also discovered my "Real Time Protection" anti-malware software is scanning every file in the cache. This has a devastating impact on performance!

Recommend that users with any anti-virus software installed should configure anti-malware settings to white-list (exclude) the location of the joblib memory cache for performance.