bug#68244: hash-table improvements

[Mattias Engdegård]

It's the hash-table's turn to be subjected to some performance tuning. The 
implementation has changed very little over the years and it's not so much a 
matter of a single big thing to fix as many small ones, so the list of changes 
is fairly long but ever little helps.

A working patch series can be found in the scratch/hash-table-perf branch. 
Although it should all be satisfactory and an improvement on what was before, 
there are a couple of details that I'd like to do better, which is why this 
hasn't been merged yet: the way shared hash_table_test structs are cached isn't 
very elegant, nor is the way we deal with Qunbound in pdumper.c.