libxpkg 0.0.53: install_dir says why it has no answer (openxlings/xlings#487)

[Sunrisepeak]

Fixes the diagnostic half of openxlings/xlings#487.

The message named an internal state

cannot get install dir for xim:libcuda-host-link@0.0.1 covers two causes
that point in opposite directions:

cause	where the reader should look
not a dependency of this package on this platform	the recipe's platform sections / the hook's branch
declared here, but the payload never landed	the dependency install

The old message pointed at neither, and the issue reasonably read it as a
path problem — hypothesising that dependency resolution was not filtering by
platform.

It was neither of those

Resolution is platform-filtered (resolver.cppm:
pkg->xpm.runtime_deps.find(platform)), and the recipe was right too —
ollama declares the CUDA sentinel only under xpm.linux. The real cause was
ollama's install hook branching on is_host("windows") when the real
distinction was linux, so macOS took the linux path and asked for a
package macOS never resolves. That half is fixed in the index
(openxlings/xim-pkgindex#532).

What changed

deps_list is what the resolver produced for this platform, so a name
missing from it is not "unresolvable" — it is "not a dependency here". The
message now says which of the two cases it is, and names the package asked
for, the platform, and what the deps here actually are:

install dir for xim:linux-only-thing@0.0.1: NOT a dependency of ollama on
this platform (macosx). Its deps here are: xim:something-else@1.0.0. A
package declared only under another platform's xpm section is never resolved
on this one -- guard the hook that asks for it with the platform it belongs
to.

Two tests, one per branch. The declared-but-absent test asserts the platform
wording is absent — describing an incomplete install as a platform
mismatch would send the reader to the same wrong place, one door down. The
Lua log goes to the process's stdout rather than HookResult::output, so
they capture it.

117 tests pass.