tests/slurm-epilog: check the cleanup status and add pathname regressions

[MaxFreedomPollard]

Fixes #1409. Test-only, no production code changes.

I saw in #1408 that this follow-up was offered to @100milliongold. If you already have it in progress, I am happy to close this one.

The three gaps from the issue:

1. mount_case now checks the cleanup's exit status. An early exit leaves every payload in place, which is what the assertions look for, so it used to pass.
2. The owned-mount-root case now also requires a sibling outside the mount to be removed, like the owned-parent and bind-mount cases. A surviving payload alone is also what a cleanup that never ran leaves behind.
3. New pathname cases for [, backslash, *, ? and a trailing newline, each next to the sibling its mishandled name would match (data1 next to data[1], datax1 next to data*1). Each asserts both halves: the payload under the mount survives and the sibling is cleaned up. They script the mount table through an awk stub that prints the mount point in mountinfo's octal-escaped form, so they run without root. They are skipped when /proc/self/mountinfo is not readable, because the script checks for it before calling awk.

Testing

Run in an Ubuntu 24.04 container: 17 passed as an unprivileged user (mount-boundary cases skipped), 21 passed as root with --privileged. ShellCheck reports nothing before or after.

To check that the new assertions fail when they should, I ran the old and new harness against mutated copies of the template:

Mutant	Old harness	New harness
pattern=$mount_point, no escaping	16 passed	[ and backslash fail (payload gone), * and ? fail (sibling spared)
mount_point=$(printf '%b' "$field")	16 passed	the trailing-newline case fails (payload gone)
exit 17 after set -ex	the owned-mount-root case passes	all four mount cases fail with "the cleanup exited 17"
rm -fr replaced by :	the owned-mount-root case passes	the owned-mount-root case fails