Re: AIO writes vs hint bits vs checksums

Here's a rebase of these patches.

The last patch, to use mprotect() to detect bad buffer accesses, needed 
some changes because of AIO. I disabled the protection completely for 
I/O worker processes, it was difficult to be more fine-grained than 
that. I didn't test it with io_uring.

I went ahead and committed the "heapam: Only set tuple's block once per 
page in pagemode" patch, because it was trivial and independent of the rest.

-- 
Heikki Linnakangas
Neon (https://neon.tech)