u3: specify madv_random for file-backed snapshot mapping

[joemfb]

Since #402 was released, there have been isolated reports of unusually high disk i/o, specifically reads. This is unexpected behavior -- our current, best guess is the kernel's readahead logic for file-backed mappings. This PR implements the standard solution to overeager readahead (MADV_RANDOM), as a draft since the diagnosis is not confirmed, and the advice may need to be behind a flag.

[mopfel-winrux]

I was able to replicate the issue using v2.7 on Digital Ocean running 14 comets with 4GB of ram and 20GB of swap.

Switching over to this PR I tried the same thing and still saw ~40MB/s reads.

[mopfel-winrux]

I ran
trace -fp [$PID] -Trfe trace=read,lseek 2> system.trace

with all the urbits running and got the below trace
https://nyc3.digitaloceanspaces.com/mopfel-winrux-s3/mopfel-winrux/2023.6.01..12.58.45-system.trace

[ashelkovnykov]

For posterity: the actual issue was #427

[joemfb]

Thanks @mopfel-winrux, @ashelkovnykov.

I think this advice is the right thing for north.bin regardless. And I'd rather not add yet-another flag to control it. But it's not pressing, and should be measured first. Leaving as a draft for now.