[gclist] Allocator algorithms.

[Hans Boehm]

I don't see a good reason not to use best fit allocation at page level.  If you
implement it with a reasonable data structure, the cost should be in the noise.
 Even the old argument that you would end up with small, unusable blocks
disappears, since one page chunks are both the smallest possible, and commonly
used.

(I used a slightly modified first-fit algorithm, largely for historical
reasons. The modification helps, but it's still not a good choice.  There is
some interaction with black listing, so fixing it is not 100% trivial, though
it shouldn't be too hard.)

Hans