add MiniCPM-o 4.5 recipe under recipes/OpenBMB

[tc-mb]

Purpose

Add a community recipe documenting how to serve openbmb/MiniCPM-o-4_5
on vLLM-Omni, following the structure of the existing
recipes/Qwen/Qwen3-Omni.md and recipes/OpenBMB/VoxCPM2.md.

The MiniCPM-o 4.5 integration itself already landed in
#3642 (model code + 3 default deploy configs + Gradio demo); this PR
only adds the recipe file + a row in the central recipes/README.md
table. No source code is touched.

What's added

• recipes/OpenBMB/MiniCPM-o-4_5.md — covers the three shipped GPU
  layouts:
  • 2-GPU default (minicpmo_4_5.yaml)
  • 3-GPU thinker TP=2 (minicpmo_4_5_3gpu.yaml)
  • 8x RTX 4090 (minicpmo_4_5_8x4090.yaml)
• Recipe also documents:
  • The stepaudio2-minicpmo talker dependency and the
    pip install 'vllm-omni[minicpmo]' install path
  • The chat_template_kwargs.use_tts_template=true gate that
    triggers speech output
  • The hf_config.version == "4.5" routing predicate that prevents
    misrouting MiniCPM-o 2.6 checkpoints into the 4.5 pipeline
• recipes/README.md — added one row to the Available Recipes table,
  inserted in alphabetical order before OpenBMB/VoxCPM2.md.

Test Plan

Documentation-only change. All relative links in the recipe resolve to
files that exist in this PR's base (verified locally against
origin/main post-rebase).

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.