add OMTD core library

[mirek190]

Summary

Adds the OMTD core library as a standalone output-modality boundary for future generated-media backends.

OMTD is the output-modality counterpart to MTMD: it keeps generated-media output routing behind a separate companion path instead of overloading input-media projector semantics.

Notes

• Scope is OMTD core only.
• Model-specific backends are intentionally left for follow-up patches.
• This branch was prepared from mirek190/llama.cpp OMTD core commit 06c714efbfc5cc544a3dbee134c1e3368adabc05, adapted onto audio.cpp:release-0.1 by omitting llama.cpp aggregate CMake files that do not exist in audio.cpp.

Validation

• Confirmed final branch diff contains only OMTD docs/source/test files.
• No audio.cpp CI has been run locally for this branch.

[0xShug0]

Currently, it’s unclear to me how OMTD interacts with the audio.cpp framework. I haven't decided whether to integrate llama.cpp for AR audio models yet. But since the changes are additive and mostly placeholders, I’ll merge them.