Model card: document gpu16 variant

Add gpu16 variant (fp16, LayerNorms in fp32) for DirectML GPUs

Parley speech models: standard (fp32) and compact (int8)

initial commit