Find out optimal splitting size for large TMs and configure translate5 and t5memory accordingly

Test what is the optimal splitting size to get good response times in concordance search even for rare words in the memory.

Feel free to play around with possible CPU/RAM configurations for testing in the cloud.

Make sure, users do never run into timeouts in the UI in translate5 and set the default MB size for the error that triggers the splitting in translate5 in t5memory with Orest to the good value.

UPD: Need to find out how to avoid duplicates, as now t5memory is capable of doing that