ci: build libusb and hidraw backends as C++ on ubuntu

[Youw]

Why

Issue #671 reported that libusb/hid.c failed to compile as C++ source. PR #811 made it C++-clean; linux/hid.c already builds as C++. Today nothing in CI exercises the C++ compilation path, so a similar regression could slip through unnoticed.

Change

Add a single step to the ubuntu-cmake job that compiles both backend hid.c files with g++ -xc++:

- name: Check C++ compile (backends as C++ sources)
  working-directory: hidapisrc
  run: |
    g++ -xc++ -Wall -Wextra -Werror -Wformat-signedness -D_GNU_SOURCE \
      -I hidapi -I libusb $(pkg-config --cflags libusb-1.0) \
      -c libusb/hid.c -o /tmp/hid_libusb_cxx.o
    g++ -xc++ -Wall -Wextra -Werror -Wformat-signedness -D_GNU_SOURCE \
      -I hidapi -I linux $(pkg-config --cflags libudev) \
      -c linux/hid.c -o /tmp/hid_hidraw_cxx.o

Flag set mirrors GNU_COMPILE_FLAGS used by the existing C build, except -pedantic is omitted. Rationale: libusb.h declares zero-size arrays (#define ZERO_SIZED_ARRAY 0) and hidapi relies on designated initialisers — both GNU extensions that g++ accepts in its default mode but rejects under strict ISO C++. The check is meant to catch real cross-language regressions (#671-style implicit conversions), not those existing-by-design extensions.

No source changes; this only adds the CI guard.

Verification

Ran the exact two commands locally against current master in WSL Ubuntu (gcc/g++ 11.4, libusb-1.0):

libusb/hid.c as C++: OK
linux/hid.c as C++: OK

Also confirmed that reverting commit e8243e1 (the #671 fix) makes the libusb command fail with the same invalid conversion errors the original issue reported — so this CI step will actually catch a regression of that fix.

Drafted with Claude Code.

[Youw]

Closing in favor of: #814