discover: fix sysfs end-of-partition calculation

[eriknordmark]

Summary

The sysfs branch of findDisks (discover.go:175) computed each partition's last-byte position as size - start + 1. That's not the right formula for anything; the sysfs size attribute is the partition's length in sectors, so the inclusive last LBA is start + size - 1. The disk-image branch a few dozen lines up (discover.go:78) already uses the correct formula:

end:    start + p.GetSize() - 1,

Impact

partitionData.end flows into:

• calculate.go:48, :53, :54, :63, :67, :69, :106, :131, :133, :134 — every free-space and grow-target computation in calculateResizes.
• run_helpers.go:150 — the shrink-target end in planResizes.

So whenever partitionresizer runs against a real block device (which routes through sysfs) rather than a disk image (which routes through the GPT-table branch), the resize plan is computed against silently wrong partition geometry. The disk-image-only test fixtures (testdata/disk.img, testdata/diskfull.img) hid this from CI.

Fix

One-line change: end := size - start + 1 → end := start + size - 1, with a comment explaining the formula matches the disk-image branch.

Extend TestFindDisks/all to assert pd.end so this regression can't slip back through.

Test plan

$ go test -run '^TestFindDisks$' ./...
=== RUN   TestFindDisks
--- PASS: TestFindDisks (0.01s)
ok  	github.com/diskfs/partitionresizer	0.014s

Signed-off-by: eriknordmark erik@zededa.com