Vulkan iGPU device selection overhaul and PCI ID API support

[0cc4m]

This changes the behaviour of the Vulkan backend to show iGPUs and dGPUs by default. The new iGPU device type (added in #15797) is used for iGPUs and the logic defaulting to dGPUs if available is handed over to the GGML API. (FYI @slaren)

Additionally, the PCI ID is returned and can be used by GGML to avoid using the same device once with CUDA and once with Vulkan, for example. This should improve the behaviour of binaries with many backends included.

I also fixed the compiler warning about missing brackets for operator precedence.

[slaren]

The device id filtering work as expected for me under Windows.

ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 2 CUDA devices:
  Device 0: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes
  Device 1: NVIDIA GeForce RTX 3080, compute capability 8.6, VMM: yes
ggml_vulkan: Found 2 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 3080 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
ggml_vulkan: 1 = NVIDIA GeForce RTX 5090 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
build: 6456 (f22656bbd) with clang version 18.1.8 for x86_64-pc-windows-msvc
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_load_from_file_impl: skipping device Vulkan0 (NVIDIA GeForce RTX 3080) with id 0000:01:00.0 - already using device CUDA1 (NVIDIA GeForce RTX 3080) with the same id
llama_model_load_from_file_impl: skipping device Vulkan1 (NVIDIA GeForce RTX 5090) with id 0000:02:00.0 - already using device CUDA0 (NVIDIA GeForce RTX 5090) with the same id
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce RTX 5090) (0000:02:00.0) - 30840 MiB free
llama_model_load_from_file_impl: using device CUDA1 (NVIDIA GeForce RTX 3080) (0000:01:00.0) - 9071 MiB free

[0cc4m]

Currently, this is just based on which backend gets loaded first, right? We might wanna set an explicit order.

[slaren]

Yes, the backends loaded first take precedence. I think it works ok at the moment, the backends are always loaded in the same order so it is predictable.