mtmd: add unlimited-ocr (converter, full MHA)

[sfallah]

Overview

• Adds baidu/Unlimited-OCR to mtmd: converts and runs the model in llama.cpp.
• Converter-only, no C++ changes.
• Registers UnlimitedOCRForCausalLM (text + mmproj). Writes the decoder sliding_window
  to the GGUF as metadata; the decoder runs full MHA and ignores it for now.
• Parity test: single-page vs the HF reference (transformers 4.46.3), full MHA, within tolerance.

Scope: conversion + full-MHA inference. Follow-up PR (stacked on this one) implements R-SWA in the decoder.

How to run

GGUF models: sabafallah/Unlimited-OCR-GGUF

build/bin/llama-mtmd-cli -hf sabafallah/Unlimited-OCR-GGUF:bf16 \
  --image tools/mtmd/test-1.jpeg -p "document parsing." \
  --chat-template deepseek-ocr \
  --temp 0 --flash-attn off --no-warmup \
  -n 4096 -c 16384 \
  --dry-multiplier 0.8 --dry-base 1.75 --dry-allowed-length 35 \
  --dry-penalty-last-n 128 --dry-sequence-breaker none

Additional information

• Unlimited-OCR is trained with R-SWA (Reference Sliding Window Attention). This PR runs
  the decoder as full MHA.

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES - I used AI assistance for code review, debugging, implementation checks, and testing. I have reviewed the submitted changes and take responsibility for the full contents of this PR.