server: add null check for context to prevent segfault on init failure

[Anirudh171202]

Overview

Fixes a segfault when llama_context creation fails.
In server-context.cpp::load_model, the return from common_init_from_params is checked for a null model but not a null context. When context creation fails, llama_n_ctx(ctx) dereferences null and crashes.
This adds a null check for ctx matching the existing model check.

Ref: #21450

Reproduction Steps:

On Apple Silicon,

 ./build/bin/llama-server -hf LiquidAI/LFM2.5-1.2B-Instruct-GGUF:Q4_K_M \
  --ctx-size 32768 \
  -ngl 99 \
  -fa auto \
  --cache-type-k q4_0 \
  --cache-type-v q4_0

It segfaulted before. Now it clean exits with an error message.

Also verified --cache-type-k q4_0 --cache-type-v q4_0 still works normally.

Requirements

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

[Anirudh171202]

@ngxson could you take a quick look? It's a 5 line null check... ggerganov already approved, just needs a second review to unblock the merge approval.