Use multiply to calculate hashtable index faster

[mmozeiko]

Change how hashtable index is calculated from 64-bit hash - instead of slow division operation use a much faster single multiply operation.

This changes hash % count to ((U128)hash * count) >> 64 calculation - but shift will be no-op, as high 64-bit value of 128-bit multiplication result will be in its own register.

It will calculate different value than hash % count but it will still be distributed reasonably uniformly in [0, count) interval.

This works because you can pretend hash value is 0.64 fixed-point value representing [0, 1) float. Thus multiplying it with integer N will give you value in [0, N) interval.

I did not change files in codeview_parse, dwarf_parse, rdi_parse, rdi_make, rdi_from_dwarf, rdi_from_pdb and linker folders, because I'm not sure in which places hash table contents depends on actual % operation result in rdi/dwarf/pdb file formats. Probably bunch of places there also can be changed to use this faster way.