ggml-blas: report memory when calling ggml_backend_blas_device_get_memory.

[pwhelan]

Implement memory information retrieval for ggml_backend_blas_device_get_memory. It uses sysinfo on linux, which I tested.

I copied the windows version from #18578. I have no desire to test it on windows and the implementation looks sane. I am also assuming the original author was more familiar with windows.

My reason for fixing this bug is to allow for light weight testing of code that requires querying the currently used memory of a device using BLAS on my laptop.

[taronaeo]

Unfortunately its not a bug. The BLAS backend uses host memory. Setting *free = 0; *total = 0; simply lets the CPU backend report the memory information instead and no code duplication is required across backend implementations.

See:

llama.cpp/src/llama.cpp

Lines 118 to 127 in f7db3f3

	// devices can return 0 bytes for free and total memory if they do not
	// have any to report. in this case, we will use the host memory as a fallback
	// fixes: https://github.com/ggml-org/llama.cpp/issues/18577
	if (free == 0 && total == 0) {
	ggml_backend_dev_t cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
	if (cpu_dev == nullptr) {
	throw std::runtime_error(format("%s: no CPU backend found", __func__));
	}
	ggml_backend_dev_memory(cpu_dev, &free, &total);
	}

[taronaeo]

Closing as not a bug. Feel free to reopen if otherwise.