ui: media attachments before text

[sfallah]

Overview

MTMD models like DeepSeek-OCR expect the media-marker before text.
mtmd-cli correctly so put media-maker before text.

related issue: #21022

related PR: #23345

Additional information

Before this change DeepSeek-OCR was not working with llama-server.
All other MTMD should still work, because they are tested in mtmd-cli with the order [image][text]/[audio][text].

Tests

I have tested through the UI with interleaving image-text-audio-text-image it works.
build/bin/llama-server -hf ggml-org/gemma-4-E2B-it-GGUF:bf16

I have also tested with DeepSeek-OCR using single image via UI and via API calling.

gemma-4-E2B-it

build/bin/llama-server -hf ggml-org/gemma-4-E2B-it-GGUF:bf16

DeepSeek-OCR

uild/bin/llama-server \
    -hf sabafallah/DeepSeek-OCR-GGUF:bf16 \
    --temp 0 -n 2048 --flash-attn off \
    --no-jinja --chat-template deepseek-ocr

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES - I have used AI investigate the UI code and help to make the changes.