Description of problem:

Whith huge memory guests (multi terabyte ram size), zero page detection takes too much time. Move the code that detects zero pages to the multifd threads.

Version-Release number of selected component (if applicable):

All.

How reproducible:

Migrate with a 4TB guest.

Actual results:

Ends having more than 74% of the time on is_zero page

Expected results:

is_zero_page don't show on the main thread profile. This is the main performance bottleneck.

Additional info: