allow fp16 in quant/dequant decompositions (#121738)
Test Plan:
```
buck2 run mode/dev-nosan mode/inplace executorch/examples/models/llama2:export_llama -- -c ~/llama/ultra_new_checkpoint.pt -p ~/llama/params.json -kv -E 8,8 -d fp16 --pt2e_quantize "xnnpack_dynamic" -2
```
Reviewed By: kirklandsign
Differential Revision: D54785950
Pull Request resolved: https://github.com/pytorch/pytorch/pull/121738
Approved by: https://github.com/jerryzh168