board: enable LTO on g431 nsh boards and fix LTO link with toolchain switch

For boards b-g431b-esc1/nsh, nucleo-g431kb/nsh and nucleo-g431rb/nsh,
the `.data` section of nuttx.elf overflows the 128 KB flash region by a
few hundred bytes on default GNU EABI builds:

    arm-none-eabi-ld: nuttx section `.data' will not fit in region `flash'
    arm-none-eabi-ld: region `flash' overflowed by 296 bytes

Enabling CONFIG_LTO_FULL=y in the corresponding defconfigs brings flash
usage back below the 128 KB limit (~93%).

Additionally, fix arch/arm/src/cmake/gcc.cmake so that when LTO is
enabled it always uses the gcc-ar / gcc-nm / gcc-ranlib wrappers, not
just when CONFIG_ARM_TOOLCHAIN_GNU_EABI is also set.  CI tooling (via
tools/testbuild.sh) configures cmake first, then flips the toolchain
choice in .config with kconfig-tweak before running `cmake --build`.
With the previous "ARM_TOOLCHAIN_GNU_EABI && !LTO_NONE" guard the
regen step would switch to plain `ar` for the .a files, even though the
linker driver picked at configure time is still arm-none-eabi-g++ and
the object files contain GCC LTO IR.  The result was a flood of
"undefined reference to `printf' / `free' / `puts' ..." link errors
when running tools/testbuild.sh -A -N -R on the *_CLANG variants.

Now that we are inside gcc.cmake the toolchain is unambiguously GCC,
so dropping the redundant CONFIG_ARM_TOOLCHAIN_GNU_EABI conjunct keeps
the LTO-aware ar wrappers in place across kconfig-tweak toolchain
switches.

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>

tools/ci/testlist/arm-09.dat

@@ -2,7 +2,7 @@
 
 /arm/stm32/nucleo-f4*,CONFIG_ARM_TOOLCHAIN_CLANG
 
-/arm/stm32/nucleo-g*,CONFIG_ARM_TOOLCHAIN_CLANG
+/arm/stm32/nucleo-g*,CONFIG_ARM_TOOLCHAIN_GNU_EABI
 
 /arm/stm32/nucleo-l*,CONFIG_ARM_TOOLCHAIN_CLANG