Mmaped pages are copy-on-write references to the zero page, and are
faulted in on the first write. Hash table resize touches most of the
newly allocated buffer, so it's faster to pre-fault all pages after
allocation, instead of having the resize loop interrupted by page
faults.

* Fix shared build
* Major default libs refactor
* Fix build with gcc_eh
* Link all libraries as a big group.
* Use global interface library as a group
* Build capnproto using our cmake
* Use only internal libunwind

It is significantly different from other uses of hash tables, and poses
the main obstacle to changing the hash table interface to the one that
can be easily supported by compound hash tables.

Make it explicitly separate, implement it only for a particular kind of
HashTable, and move this implementation to the ReverseIndex module.

And tests;

* Function to check if given token is present in a string;
* Special case for hasToken to 'tokenbf_v1' index;
* Test cases for hasToken()
* Test case for hasToken() + 'tokenbf_v1' integration