Made the example builds work with LLVM, and fixed four that were broken on Linux

[fdesbiens]

Makes the example builds work with LLVM, and fixes four that could not run on Linux at all.

Why the gnu examples rather than the ac6 ones

The ac6 examples were evaluated as the place to add LLVM support. Measurement shows the gnu examples are already the LLVM path, and the ac6 ones would have to be rebuilt from scratch:

	gnu example	ac6 example
Linker input	GNU ld script, which ATfE consumes directly	scatter file, which ATfE does not
Startup	crt0.S and vectors, in tree	depends on __main and Image$$ARM_LIB_STACKHEAP$$ZI$$Limit from armclib and armlink
Build driver	shell and batch scripts	none; 83 Arm DS project files

Porting the ac6 examples means writing GNU linker scripts, replacing the startup path and writing build drivers, which reproduces the gnu example that already exists for every one of those cores. This change teaches the gnu scripts to drive either toolchain instead.

Toolchain selection

Each script picks its toolchain from TOOLCHAIN, defaulting to gnu, so every existing invocation behaves exactly as before:

TOOLCHAIN=atfe ATFE_CLANG=/path/to/clang ./build_threadx_sample.sh

The differences are four values: the compiler, the target triple, -Wl,--entry= instead of the driver's -e, and -lsemihost instead of --specs=nosys.specs. That last one matters: --specs= is a GCC spec-file mechanism with no LLVM equivalent, and picolibc leaves _exit to the application, so using the toolchain's own semihosting library avoids adding a syscall stub source to every example and suits an example that runs on a model.

Parameterised, not duplicated. Duplicate build scripts drift the first time one side is edited, which is the failure #590, #591 and #592 addressed.

Four scripts that could not run on Linux

The A-profile sample scripts referenced filenames that do not exist:

Script	References	On disk
A5, A9	MP_GIC.s	MP_GIC.S
A5, A7, A8, A9	MP_PrivateTimer.s	MP_PrivateTimer.S
A5, A7, A8, A9	V7.s, and V7.o on the link line	v7.s

Case-insensitive on Windows, fatal on Linux. Corrected to match the files, which is why the Cortex-A5, A7, A8 and A9 examples now build on Linux where before they could not.

Results: every example, both toolchains

Core	GNU	ATfE
Cortex-M3	ok 21,456 B	ok 22,650 B
Cortex-M4	ok 21,624 B	ok 22,810 B
Cortex-M7	ok 21,540 B	ok 22,842 B
Cortex-A5	ok 42,100 B	ok 40,098 B
Cortex-A7	ok 41,820 B	ok 39,326 B
Cortex-A8	ok 41,860 B	ok 39,906 B
Cortex-A9	ok 42,124 B	ok 40,162 B
Cortex-A12, A15, A17	ok 41,768 B	fails
Cortex-M0, arm9, arm11, Cortex-R4, R5	fails	fails

check_clang.sh now links the examples as well as compiling the sources, since compiling proves the sources parse while only linking exercises entry symbols, linker scripts and the C library together. The eight that do not link are listed in the script with their reasons, so the gaps stay visible instead of being silently skipped.

Two defects this surfaced, neither fixed here

The Cortex-M0 example cannot link with any toolchain. cortexm0_crt0.S references __text_load_start__, __text_start__ and __text_end__, and its linker script never defines them; the Cortex-M4 script defines the equivalents. It fails identically on dev before this change, so it is pre-existing rather than introduced here.

Cortex-A12, A15 and A17 fail only under LLVM, for a specific reason: their link line omits -nostartfiles, so the toolchain's own crt0.o is linked, and ATfE's wants picolibc's __data_start, __data_source, __data_size and __bss_size, which their linker script does not define. Defining those symbols is the proper fix, but they drive a data-copy loop that cannot be validated without executing the image, so it is left to separate work rather than guessed at.

The arm9, arm11, Cortex-R4 and Cortex-R5 failures are environmental: those scripts need newlib multilib variants that are not present in every GNU toolchain packaging. They also fail on dev.

Not in scope

The .bat equivalents are untouched. ATfE on Windows has not been validated, and changing those scripts without it would imply support that has not been demonstrated.