Support Qwen2_5_VLForConditionalGeneration

[csabakecskemeti]

Make sure to read the contributing guidelines before submitting a PR

Based on this discussion:
#11483 (comment)

for the language part we can use the same architecture as Qwen2VLForConditionalGeneration.

Additionally HimariO has the update for the qwen2_vl_surgery.py to support --model_type "qwen2.5vl"
https://github.com/HimariO/llama.cpp.qwen2vl/tree/qwen25-vl
python examples/llava/qwen2_vl_surgery.py <path_to_model> --data_type fp32 --model_type "qwen2.5vl"

Tested working, examples:
DevQuasar/Qwen.Qwen2.5-VL-7B-Instruct-GGUF

build/bin/llama-qwen2vl-cli -m ~/Downloads/Qwen.Qwen2.5-VL-3B-Instruct.Q4_K_M.gguf --mmproj qwen.qwen2.5-vl-3b-instruct-vision.gguf -p "Describe this image." --image ~/Downloads/flower.jpg

encode_image_with_clip: image encoded in  3824.77 ms by CLIP (   54.64 ms per image patch)

The image depicts a flower with a unique and surreal appearance. The flower has a predominantly purple hue with a gradient effect, transitioning from deep purple to lighter shades as it moves towards the center. The petals are somewhat elongated and have a wavy, fluid texture, giving the flower a sense of movement or transformation. The center of the flower is a bright, electric blue, which contrasts sharply with the surrounding petals. The flower is set against a background that appears to be a blurred, soft green, which complements the purple and blue tones of the flower. Overall, the image has an artistic and somewhat otherworldly quality, with a focus on the vivid colors and intricate details of the flower.
llama_perf_context_print:        load time =   11328.57 ms