server : properly handle null llama_context

[fairydreaming]

Overview

When researching #25851 I noticed that llama-server does not handle case where llama_context creation failed and ctx_tgt is null, instead it crashes trying to access null context pointer. This PR adds a context nullity check to handle this case.

Additional information

This is purely cosmetic improvement - instead of:

./bin/llama-server -m ~/ggufs/DeepSeek-V4-Flash.gguf -c 262144 -ctk q8_0 -ctv q8_0 -lv 4 -fit off -fa off -cmoe
...
0.10.162.893 I load_tensors: offloading output layer to GPU
0.10.162.901 I load_tensors: offloading 42 repeating layers to GPU
0.10.162.902 I load_tensors: offloaded 44/44 layers to GPU
0.10.162.917 I load_tensors:   CPU_Mapped model buffer size = 141362.00 MiB
0.10.162.920 I load_tensors:        CUDA0 model buffer size =  7766.93 MiB
0.10.683.593 I cmn  common_init_: added <｜end▁of▁sentence｜> logit bias = -inf
0.10.692.115 E llama_init_from_model: V cache quantization requires flash_attn
0.10.692.118 E cmn  common_init_: failed to create context with model '/home/phm/ggufs/DeepSeek-V4-Flash.gguf'
0.10.692.156 Segmentation fault (core dumped)

we will have:

./bin/llama-server -m ~/ggufs/DeepSeek-V4-Flash.gguf -c 262144 -ctk q8_0 -ctv q8_0 -lv 4 -fit off -fa off -cmoe
...
0.09.491.879 I load_tensors: offloading output layer to GPU
0.09.491.885 I load_tensors: offloading 42 repeating layers to GPU
0.09.491.885 I load_tensors: offloaded 44/44 layers to GPU
0.09.491.892 I load_tensors:   CPU_Mapped model buffer size = 141362.00 MiB
0.09.491.893 I load_tensors:        CUDA0 model buffer size =  7766.93 MiB
0.10.004.591 I cmn  common_init_: added <｜end▁of▁sentence｜> logit bias = -inf
0.10.005.510 E llama_init_from_model: V cache quantization requires flash_attn
0.10.005.512 E cmn  common_init_: failed to create context with model '/home/phm/ggufs/DeepSeek-V4-Flash.gguf'
0.10.005.515 E cmn  common_init_: failed to create context with model '/home/phm/ggufs/DeepSeek-V4-Flash.gguf'
0.10.005.520 E srv    load_model: failed to create_context with model '/home/phm/ggufs/DeepSeek-V4-Flash.gguf'
0.10.005.522 I srv    operator(): operator(): cleaning up before exit...
0.10.016.392 E srv  llama_server: exiting due to model loading error

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: NO