setup: detect any recompiled BIOS stem and stop the first-run setup loop

[Alexbeav]

Summary

Focused port of the fork's fix/setup-host-bios-deadlock feature onto current
master, prepared as a single squashed change because the fork line and
upstream share no merge base (patch-port route, decided 2026-09-10).

The setup host only accepted OpenBIOS and SCPH1001 style stems. A kit that
pins another recompiled retail BIOS stem could loop through first-run setup
without ever finishing. This change:

• accepts any recompiled BIOS stem the codegen host produced
  (runtime/include/psx_bios_known_images.h),
• checks the setup host before validating a chosen BIOS (runtime/src/main.cpp),
• makes the first-run loop breaker explain its cause instead of looping,
• decodes CLI subprocess output as UTF-8 rather than the ANSI codepage
  (psxrecomp_cli.py),
• registers the two new CTest entries so they actually run
  (recompiler/CMakeLists.txt),
• names the recursive submodule init in the rbengine preflight
  (runtime/runtime.cmake).

Port notes

• Two conflicts against current master, both resolved: the CTest
  registrations (the two names exist only in this change) and the recomp-ui
  guidance text (takes the new wording).
• Cubic review P3 finding on the fork PR fixed here as well: the sibling
  PSX_NETPLAY FATAL_ERROR at runtime/runtime.cmake now names
  git submodule update --init --recursive, matching the corrected message.

Evidence and limits

• python -m py_compile passes for psxrecomp_cli.py and the new
  runtime/tests/test_codegen_host_bios_stems.py.
• The new test skips in a tree without RECOMP_UI_ROOT; a full run needs a
  tree with recomp-ui present and a built codegen host. No full runtime
  build, native run, or retail route is claimed in this PR.
• The fork-side review (cubic) ran on the pre-port branch; findings were
  addressed there and are carried in this port.

Developed with AI assistance (Cline session, 2026-09-10); the changes were
validated as described above before opening this PR.