Falcon optimization

[libinta]

What does this PR do?

1. add use_flash_attentiong, flash_attention_recompute, flash_attention_causal_mask
2. add mark step per decoder
3. add fusedsdpa fp8

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[ssarkar2]

measurement:
QUANT_CONFIG=./quantization_config/maxabs_measure_include_outputs.json python run_generation.py --model_name_or_path mistralai/Mistral-7B-Instruct-v0.2 --attn_softmax_bf16 --use_hpu_graphs --trim_logits --use_kv_cache --reuse_cache --bf16 --batch_size 1 --warm 0 --n_iter 1 --flash_attention_recompute --flash_attention_causal_mask --use_flash_attention

actual run:
QUANT_CONFIG=./quantization_config/maxabs_quant.json python run_generation.py --model_name_or_path tiiuae/falcon-40b --use_hpu_graphs --use_kv_cache --limit_hpu_graphs --bucket_size 128 --max_new_tokens 2048 --batch_size 4 --bf16 --reuse_cache --bucket_internal --fp8 --max_input_tokens 2048 --warm 2 --n_iter 2 --flash_attention_recompute --flash_attention_causal_mask --use_flash_attention

for 2048->2048, falcon 40b

with bs8:
without PR: 185 tps with 84.2 GB memory
with PR: 187 tps with 72.9 GB

with BS12:
without PR: OOM
with PR: 219 tps with 90.63 HB memory