register openkal-llvm-runtime 0.15.1

[Sunrisepeak]

Correction, added after merge. The title and the text below say
"registration only" and "pins.toml still names 0.15.0". That is not what
this commit did. An uncommitted pins.toml edit from another branch
followed a git checkout and was swept in by git add -A, so this commit
also moved runtime from 0.15.0 to 0.15.1 — the separation this
repository keeps between registering a version and pinning it was broken
here by accident.

The consequence was measured rather than assumed. On this PR branch the
compat workflow ran with the moved pin while 0.15.1 was not yet on main,
so xlings could not install it:

error: xlings install_packages failed (exit 1) for
       'openkal-llvm-runtime@0.15.1' with an unknown index-repo configuration

55 of 60 cells failed, and the workflow still reported success, because
the ratchet is not a required check. .xpkgindex/openkal-compat.json was not
touched, so nothing false was published; what was lost is the measurement
that should have accompanied the pin. It is being taken now that 0.15.1 is
registered and published.

This is precisely the failure the two-PR rule exists to prevent, arriving
through a mechanism the rule does not cover: not a wrong decision about
ordering, but an uncommitted file travelling between branches.

---

Registration only — tests/openkal/pins.toml still names 0.15.0, and moving that pin is a separate change.

0.15.1 closes the chain that carries one Windows fix: it pins openkal-musl 0.19.1, which pins openkal-windows 0.10.1, which makes a file opened for writing describable.

kal_fs_file_info answers through NtQueryInformationFile(FileBasicInformation), which requires FILE_READ_ATTRIBUTES, and FILE_GENERIC_WRITE does not carry it — so a write-only handle held every right except the one needed to ask it about itself, and fstat on a descriptor opened a line earlier returned EACCES. libarchive performs that pair when opening an archive for output.

Measured on three real-Windows conformance jobs: with the fix the query answers 0, without it 6 (kal_err_access). Wine answers 0 either way and is recorded there as not being a second opinion.

Both urls were fetched, not merely linted for shape: 200 and 15629730 bytes on each host, and the CN copy is byte-for-byte identical to the GitHub archive.