feat(backend): add moss-tts-cpp text-to-speech backend

[localai-bot]

Summary

Adds a new moss-tts-cpp Go + purego backend wrapping moss-tts.cpp, the C++/ggml inference port of the OpenMOSS MOSS-TTS family. It serves text-to-speech for the MOSS-TTS-Local v1.5 model (a GPT-J local transformer decoded through MOSS-Audio-Tokenizer-v2), producing 48 kHz stereo audio with optional reference-audio voice cloning, entirely on ggml with no Python at inference.

It mirrors the existing qwen3-tts-cpp backend: dlopen the static-ggml shared library, bind the moss-tts.cpp C-API via purego, and serve the gRPC TTS method. A thin C shim holds the pipeline handle and copies engine PCM into a Go-freeable buffer.

GGUFs are published at mudler/MOSS-TTS-Local-Transformer-v1.5-GGUF (f16 + q8_0 + codec + tokenizer).

What is included

• backend/go/moss-tts-cpp/ — the backend (main.go purego wiring, gomossttscpp.go gRPC TTS impl + 3-GGUF resolution, cpp/ C shim, audio.go stereo WAV writer, options.go, CMakeLists.txt static-ggml link, Makefile pinning moss-tts.cpp, run.sh/package.sh/test.sh, README.md, tests).
• CI registration — .github/backend-matrix.yml (CPU, CUDA 12/13, Intel SYCL f16/f32, Vulkan, ROCm, NVIDIA L4T, and Darwin metal), backend/index.yaml metas + image entries, root Makefile build targets, and the scripts/changed-backends.js path mapping. Tag-suffixes verified consistent between the matrix and index.yaml.

Test plan

• gofmt clean, go build ./... in backend/go/moss-tts-cpp/, go vet clean (same unsafe.Pointer nolint pattern as qwen3-tts-cpp).
• changed-backends.js resolves the backend to backend/go/moss-tts-cpp/ (Linux + Darwin).
• matrix/index.yaml shape parity with qwen3-tts-cpp; every image tag maps to a real build tag.
• CI backend build + a smoke synth on the published GGUFs.

Notes

This is an AI-assisted contribution (Assisted-by: trailer per the project policy; the submitter owns and must review it). The local pre-commit coverage gate was skipped for this commit because this dev box currently has :9090 bound by another process; the change is coverage-neutral (no core//pkg/ Go code, only a new backend module + registration data), so CI will run the real gate on clean infra. DCO sign-off pending from the human submitter.