gh-129201: Use prefetch in GC mark alive phase.

[nascheme]

This PR implements prefetching as suggested in the issue. The benchmarks that are part of pyperformance generally don't show much difference since they don't use enough memory to make prefetch effective.

• Issue: Use "prefetch" CPU instructions during the marking phase of the GC #129201

The benchmark results below come from my own workstation (AMD Ryzen 5 7600X, Linux, GCC 12.2.0). I'm using --enable-optimizations but LTO is turned off. The "prefetch off" means the prefetch CPU instruction is omitted, otherwise code is the same.

There is something bad happening with the default build GC is gc.disable() is not used when creating a big data structure. When enabled, the "gc big" benckmark takes 5x as long. My guess is that the generation 0 collections are shuffling the objects in the gc next/prev linked list but that's only a guess.

Source code for the two "big" benchmarks. These create a fairly large object graph and then call gc.collect() to time it.

gc big tree
gc big

The "bm_gc_collect" benchmark was taken from pyperformance and the constants adjusted: CYCLES = 100_000 and LINKS = 40.

The "prefetch (7f756eb0)" code branch is essentially the same as this PR (1b4e8c3). I just rebased it on the current main and removed some dead code.

code branch	build	prefetch on	benchmark	time [ms]
benchmark: bm_async_tree_io_tg
prefetch (7f756eb0)	nogil	yes	bm_async_tree_io_tg	6,752
prefetch (7f756eb0)	nogil	no	bm_async_tree_io_tg	6,964
main (3829104)	nogil	-	bm_async_tree_io_tg	6,960
main (3829104)	default	-	bm_async_tree_io_tg	7,323
benchmark: bm_gc_collect
prefetch (7f756eb0)	nogil	yes	bm_gc_collect	1,418
prefetch (7f756eb0)	nogil	no	bm_gc_collect	1,517
main (3829104)	nogil	-	bm_gc_collect	1,473
main (3829104)	default	-	bm_gc_collect	20,966
benchmark: gc big
prefetch (7f756eb0)	nogil	yes	gc big	587
prefetch (7f756eb0)	nogil	no	gc big	1,064
main (3829104)	nogil	-	gc big	963
main (3829104)	default	-	gc big	5,785
benchmark: gc big tree
prefetch (7f756eb0)	nogil	yes	gc big tree	363
prefetch (7f756eb0)	nogil	no	gc big tree	519
main (3829104)	nogil	-	gc big tree	723
main (3829104)	default	-	gc big tree	3,443
benchmark: gc big (gc disabled)
main (3829104)	default	-	gc big (gc disabled)	641