feat(mcpp.plugins): 0.5.0, an island's entry points in the module's namespace

[Sunrisepeak]

Publishes mcpp-plugins v0.5.0, merged as 992a1b4 with a green run on main.

What the release changes

docs/42 states one rule for both lanes: the module name and the namespace are one identifier path. The shader lane followed it; mcpp.tools.island did not, and emitted every entry point at global scope. A root's directories now extend the namespace exactly as a payload tree's do:

src/backends/cuda/image/blur.cu   myapp_blur   ->  myapp::kernels::image::myapp_blur

options::roots and options::layout_root replace the file list; options::strip_prefix adds a short spelling beside the authored name, which stays canonical because it is the symbol.

Why latest moves even though the release is breaking

Every published consumer pins an exact version -- a bare version is read as a choice, not a range -- so nothing resolving today is affected. The three examples in mcpp that use the generator move to 0.5.0 in the same batch, and their pull request is opened after this one lands because they resolve this package through the index.

The floor does not move. 0.5.0 changes what the package generates, not what it asks the engine for, so it stays at mcpp 2026.9.8.1.

Checked before opening

check	result
loadfile(..., 't') syntax	ok
check_cross_package_refs, check_platform_version_parity, check_duplicate_versions	ok
mcpp xpkg parse	versions linux 0.5.0, 0.4.0, ... and parse OK
exactly one ["latest"] per platform table, all three at 0.5.0	3 rows, 3 tables
the CN asset downloads and matches	ec11aa99... equals the GitHub source archive's sha256

That last one is worth stating: gtc has reported uploaded for an object that then answered 404, so the asset was fetched back and hashed rather than trusted.