fix(disk/aix): use statfs for UsageWithContext on AIX

[pgimalac]

Update UsageWithContext on AIX to use the same implementation as other Unix OSes, based on the statfs syscall.

• Aligns implementation with other OSes
• Avoids running commands (and parsing the output) in the CGO_ENABLED=0 case
• Makes UsageWithContext work with arbitrary paths. The old implementations (both CGO and no-CGO) only work when path is exactly a mountpoint (fails with mountpoint ... not found otherwise), while statfs works properly no matter what

I confirmed getting the exact same output for real mountpoints, and it now works for non-mountpoints.

[shirou]

Merging this — thank you so much! Nice cleanup, and it's also a return to what we had: until May 2024 disk_unix.go was tagged ... || (aix && !cgo) and AIX nocgo used this same statfs path; #1651 swapped in the df -v parser with no stated reason.

One thing I'd like on the record here, if you don't mind answering after the fact: AIX's statfs has both f_bsize and f_fsize, and the shared code scales the block counts by Bsize. On JFS2 the two are identical, but classic JFS has fragments — does Blocks * Bsize still give the right total there? I couldn't reach IBM's docs to check either way. "JFS2 is what matters in practice" is a perfectly good answer — I just want it written down, so whoever hits a wrong Total on AIX later has a place to start. Whichever build and filesystem types you checked against would be useful in the same breath.

Two things I'll take care of separately, nothing for you to do:

• The removed cgo UsageWithContext fell back to "unknown" for unmapped vfstypes and PartitionsWithContext still does, so a cgo build now reports "unknown" from Partitions() and "" from Usage(). "" matches the other platforms, so I'll leave getFsType as is and align Partitions() on its own.
• On nocgo the values shift slightly: the old parser read %Used/%Iused with ParseInt (always whole numbers) and took Free from df's column rather than Bavail. The new ones are better — Used/(Used+Free) is the project-wide convention from #562 — I'll note the change in the release notes.

No tests needed; we have no AIX runner. For context, our lint matrix also only builds AIX with CGO_ENABLED=0 (the cgo row is commented out with a # FIXME), so CI doesn't cover the side this PR actually changes. That's our gap, not yours.