Optimising compactify_tuples()

Hi,

With [1] applied so that you can get crash recovery to be CPU bound
with a pgbench workload, we spend an awful lot of time in qsort(),
called from compactify_tuples(). I tried replacing that with a
specialised sort, and I got my test crash recovery time from ~55.5s
down to ~49.5s quite consistently.

I've attached a draft patch. The sort_utils.h thing (which I've
proposed before in another context where it didn't turn out to be
needed) probably needs better naming, and a few more parameterisations
so that it could entirely replace the existing copies of the algorithm
rather than adding yet one more. The header also contains some more
related algorithms that don't have a user right now; maybe I should
remove them.

While writing this email, I checked the archives and discovered that a
couple of other people have complained about this hot spot before and
proposed faster sorts already[2][3], and then there was a wide ranging
discussion of various options which ultimately seemed to conclude that
we should do what I'm now proposing ... and then it stalled. The
present work is independent; I wrote this for some other sorting
problem, and then tried it out here when perf told me that it was the
next thing to fix to make recovery go faster. So I guess what I'm
really providing here is the compelling workload and numbers that were
perhaps missing from that earlier thread, but I'm open to other
solutions too.

[1] https://commitfest.postgresql.org/29/2669/
[2] https://www.postgresql.org/message-id/flat/3c6ff1d3a2ff429ee80d0031e6c69cb7%40postgrespro.ru
[3] https://www.postgresql.org/message-id/flat/546B89DE.7030906%40vmware.com