[feat] Add GGUF conversion and inference support for BitNet embedding 270m (Gemma3)

[isHuangXin]

• Add convert-bitnet-embedding-270m-to-gguf.py for Gemma3-based 270m models
• Support f32, f16, and I2_S ternary quantization output types
• Add AVX512BW SIMD paths for I2_S dot product in ggml-bitnet-mad.cpp
• Add immintrin.h include and bitnet-lut-kernels.h guard in ggml-bitnet-lut.cpp
• Add documentation for Gemma3 GGUF conversion implementation
• Update llama.cpp submodule with Gemma3 architecture support

[isHuangXin]

@microsoft-github-policy-service agree