An empty-string path operand resolves to the cwd, so every command reports 'Is a directory' for ''

[davydog187]

An empty-string path operand resolves to the cwd, so every command reports Is a directory for ''.

$ sort ''
sort: read failed: : Is a directory          # gsort '': "cannot read: '': No such file or directory"

$ cat ''
cat: : Is a directory                        # gcat '': "cannot open '' for reading: No such file..."

FS.resolve_path(cwd, "") returns the cwd, so the empty operand becomes a directory that exists, and the
command faithfully reports what the filesystem told it. The command is not at fault — the path layer is.

POSIX is explicit that an empty pathname resolves to nothing: it is ENOENT, never the current
directory. Every command taking a file operand inherits the wrong answer, and it is the kind of input
that shows up from an unquoted empty variable (sort "$f" with f unset), so it is reachable without
anyone typing ''.

The fix belongs in FS.resolve_path/2 — return {:error, :enoent} for an empty component rather than
collapsing to the base. Worth checking what resolve_path/2 does with the other degenerate inputs at the
same time (a bare /, repeated slashes, a trailing .), since they go through the same normalisation.

Note that cp/mv/ln/tee/touch/sed -i and the redirect path now route destination spellings
through FS.check_directory_spelling/3 (PR #75), so an empty destination may need the same treatment
as an empty source.

---

Pre-existing at origin/main (82ee9e5) — cat '' already misreports there, before any of the current
PRs. Deliberately deferred out of PR #72 and PR #74 because the fix is in the path layer and would have
conflicted with both.

Found by an independent verification pass while working #68 and #70.