Handle active swap partitions when opening an existing disk

[VykosMolt]

Fixes #4235

Install once with a normal swap partition, then open the installer again on
that disk. Swap shows up in the partition list as if it were mounted at a
folder called [SWAP], and tearing the layout down fails.

Both come from the same place. When a partition is in use as swap, lsblk
prints [SWAP] where it would normally print a mountpoint, in both
mountpoint and mountpoints. LsblkInfo turned that into Path('[SWAP]'),
so as far as the rest of the code was concerned the partition was mounted at a
folder of that name. umount() then saw a non-empty list of mountpoints,
skipped the early return it has for unmounted partitions, and ran
umount -R '[SWAP]'.

Two changes.

LsblkInfo no longer treats [SWAP] as a path. It is matched exactly rather
than stripping anything in brackets, because [SWAP] is the only bracketed
value lsblk prints and a real folder is allowed brackets in its name. That
also keeps it out of the partition list, which I think matches what you meant
by "could simply be skipped".

umount_all_existing() now sends swap partitions to a new swapoff() helper
next to swapon(), rather than trying to unmount them. Swap is not mounted at
a folder, so there is nothing to unmount; it has to be switched off or the
partition stays busy and the disk cannot be repartitioned.

swapoff itself errors if you point it at something that is not currently in
use as swap, so the helper asks swapon --show what is in use before doing
anything. That makes it safe to call on a partition whose swap is already off,
which matters because the installer cannot tell the two apart. Both paths are
resolved before comparing, since swap can be switched on through a link like
/dev/disk/by-uuid/... while swapon reports the device it points at.

I took that list from swapon rather than from lsblk on purpose. A swap area is
not always a block device, and get_lsblk_info() raises on a swap file:

get_lsblk_info(/swapfile)  -> DiskError: Failed to read disk "/swapfile" with lsblk
get_lsblk_info(/dev/zram0) -> ok

Going through swapon means the helper works for partitions, encrypted and LVM
devices, and swap files alike, so it should be reusable for the teardown work
in #4536 and for the swap file in #4729 without either needing its own copy.
Real failures raise DiskError, matching swapon(), so nothing is swallowed.

Nine tests cover both fields carrying [SWAP], swap that is not currently on,
ordinary mountpoints, a real mountpoint with brackets in it, the symlink case,
and both failure paths. Seven of them fail on master.

One known gap: LUKS-encrypted swap still will not tear down, because
cryptsetup close finds the mapper busy. That is broken on master today too
and it needs a LUKS setup to test properly, so I have left it rather than
guess at it.

One limitation worth stating: swapon --show --raw prints names unquoted, so a
swap path containing whitespace would be ambiguous. Nothing archinstall creates
looks like that, but it is a real bound rather than a non-issue.

Claude Opus was used for codebase archaeology, implementation assistance, and
test review. I reviewed and tested the changes myself.

[svartkanin]

This seems to be a purely AI generated PR, we don't have a AI policy (yet) so I'll leave these comments for it

• Please write a human description
• Please remove AI comments as they rarely add any value and just add verbosity
• Please test the code and review it yourself before submitting