mcpp 2026.9.26.1 as the measured client, and the graph's own Linux targets in the openkal measurement - #469

…rgets in the openkal measurement

Pins move, with no registration in this change (openkal-musl 0.19.2 and
openkal-llvm-runtime 0.15.2 were registered in #467 and #468 and resolved from
the published index before anything here named them):

- validate.yml and openkal-compat.yml install mcpp 2026.9.26.1, and
  index.toml's latest_mcpp names it. min_mcpp does not move: the release
  changes no descriptor grammar.
- tests/openkal/pins.toml: runtime 0.15.2 (which pins openkal-musl 0.19.2
  exactly), mcpp 2026.9.26.1, and two targets, x86_64-linux-musl and
  aarch64-linux-musl. From mcpp 2026.9.26.1 a link over the graph's C library
  searches no host library directory (mcpp-community/mcpp#696); before it, a
  member's `-lm` on x86_64-linux-musl linked the host's glibc objects without
  a diagnostic. Neither target was measured here, which is why neither showed.
- compat.py: a Linux image of the host's own architecture runs without a
  runner (`host_executes`), so x86_64-linux-musl is measured as run rather
  than `--no-run`; aarch64-linux-musl runs under qemu-aarch64, which the
  workflow now installs. Three selftest cases state the rule.
- compat.libaio, compat.libdrm and compat.libinput: the comments that blamed
  `c_standard = "gnu11"` itself now name the defect (a dependency's own C
  standard was dropped up to mcpp 2026.9.25.1, mcpp-community/mcpp#695) and
  keep the `-D` spellings, which work under every engine this index admits.