Multi Model Decoder Modification

[chilukam-qti]

Contains Following Modifications:

1. KV Cache Updated after inference of model conatins KV Cache as output
   1. KV cache Indices added to pipelines containing kv cache as inputs or outputs (previously done only for kv cache as input)
   2. KV cache updates post inference of model containing kv cache as output
2. Memcopy of embeddings considers float datatype and uint16 datatype for count in copy_n
3. Generalised pixel_values tensor name based on genai_config
4. Optimized Sliding Window based KVCache copy from present to past by copying only cache for seqlen instead of entire context length