tests : refactor test-backend-sampler

[ggerganov]

cont #17004
fix #18700 (comment)

The new backend sampler tests have been clogging the ggml-ci recently by adding almost ~10 extra min runtime to some workflows. Optimize and deslop the implementation:

• Avoid loading the model on every test
• Better names
• Better RAII
• Remove redundant CMake test declarations
• Add device "auto" - used by default
• Use smaller Q4_0 model instead of F16

[CISC]

@reeselevine Broke WebGPU CI since greedy test actively fails now:

llama.cpp/tests/test-backend-sampler.cpp

Lines 1124 to 1134 in 1ffc529

	static void run_tests(const std::vector<const backend_test_case *> & tests, const test_params & args) {
	for (const auto & test : tests) {
	fprintf(stderr, "\n=== %s ===\n", test->name.c_str());
	try {
	test->fn(args);
	} catch (const std::exception & e) {
	fprintf(stderr, "Error running test '%s': %s\n", test->name.c_str(), e.what());
	exit(EXIT_FAILURE);
	}
	}
	}

https://github.com/ggml-org/llama.cpp/actions/runs/20897547250/job/60038427586#step:4:21395

[reeselevine]

Looks like we need to implement support for ARGMAX, I'll prioritize this!