bug#68244: hash-table improvements

[Stefan Monnier]

> I've pushed two new changes: a correction to the GC accounting for the
> ancillary hash-table vectors, and a rather more interesting change to the
> hash table range reduction. It now uses a Knuth multiplicative method
> instead of the expensive remainder, so the index is now always a power of
> 2 in size.

The change gives good results for small tables but less so for big ones.
I don't have a good intuition for why that would be: none of the
operations directly involved seem to be more costly for large tables, so
my best guess is that it leads to more collisions somehow, tho I don't
have a good idea about why that would be.


        Stefan