[not-for-merge] Thread compiler xvector example and few optimizations

[gorinars]

Example of running xvector extractor in multiple threads (includes e4da298 that makes it safe)

Some features include:

• --num-threads
• extra argument cache_rxfilename that allows to store cache and re-use it in extractor
• --chunk-sampling-rate - dummy implementation of frame rounding to make segments less diferse and make code spend less time on compilation (not tested properly - better use default)

[danpovey]

This is related to #2288.

@gorinars, would you mind fixing the conflicts?

As previously discussed, I will close this PR but I wanted it on github for discoverability, since this might come in useful in future.

[gorinars]

I was not sure if it was better to keep your commit e4da298 in this PR. Basically without it the thing will not work properly, so I guess it is better to keep it and later when you merge your part the diff will be not so huge

[danpovey]

closing but please keep that branch up for reference.