fix: skip main_gpu validation when no gpus are available

[Dev-iL]

Overview

Setting --split-mode none on a CPU-only build causes model loading to fail (see trace below), because main_gpu defaults to 0 and the bounds check fired against an empty device list. The accompanying warning already states that split mode should have no effect without GPU support - so this PR makes it so we skip the GPU filtering block entirely.

Additional information

[52521] warning: llama.cpp was compiled without support for GPU offload. Setting the split mode has no effect.
[52521] 0.00.024.764 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
[52521] 0.00.024.768 I device_info:
[52521] 0.00.024.788 I   - CPU     : AMD EPYC 9745 128-Core Processor (1030363 MiB, 1030363 MiB free)
[52521] 0.00.024.842 I system_info: n_threads = 256 (n_threads_batch = 256) / 512 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
[52521] 0.00.024.847 I srv          main: n_parallel is set to auto, using n_parallel = 4 and kv_unified = true
[52521] 0.00.024.889 I srv          init: running without SSL
[52521] 0.00.024.916 I srv          init: using 511 threads for HTTP server
[52521] 0.00.025.001 W srv          main: -----------------
[52521] 0.00.025.003 W srv          main: Built-in tools are enabled, do not expose server to untrusted environments
[52521] 0.00.025.003 W srv          main: This feature is EXPERIMENTAL and may be changed in the future
[52521] 0.00.025.003 W srv          main: -----------------
[52521] 0.00.025.006 I srv         start: binding port with default address family
[52521] 0.00.026.184 I srv          main: loading model
[52521] 0.00.026.193 I srv    load_model: loading model '/llm-models/Qwen3-Coder-Next_UD-Q6_K/Qwen3-Coder-Next-UD-Q6_K-00001-of-00003.gguf'
[52521] 0.00.026.232 I common_init_result: fitting params to device memory ...
[52521] 0.00.026.234 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
[52521] 0.00.076.447 E llama_prepare_model_devices: invalid value for main_gpu: 0 (available devices: 0)
[52521] 0.00.077.830 E llama_model_load_from_file_impl: failed to load model
[52521] 0.00.077.880 E common_fit_params: encountered an error while trying to fit params to free device memory: failed to load model
[52521] 0.00.125.935 E llama_prepare_model_devices: invalid value for main_gpu: 0 (available devices: 0)
[52521] 0.00.127.042 E llama_model_load_from_file_impl: failed to load model
[52521] 0.00.127.047 E common_init_from_params: failed to load model '/llm-models/Qwen3-Coder-Next_UD-Q6_K/Qwen3-Coder-Next-UD-Q6_K-00001-of-00003.gguf'
[52521] 0.00.127.050 E srv    load_model: failed to load model, '/llm-models/Qwen3-Coder-Next_UD-Q6_K/Qwen3-Coder-Next-UD-Q6_K-00001-of-00003.gguf'
[52521] 0.00.127.057 I srv    operator(): operator(): cleaning up before exit...
[52521] 0.00.138.291 E srv          main: exiting due to model loading error

Requirements

• I have read and agree with the contributing guidelines - YES
• AI usage disclosure: YES - was used to diagnose and verify the correctness of the fix.