Moved export_meta_llama_bin.py to new export.py

[atamurad]

Follow up to #326

Summary:

• Tested updated export.py with Meta LLama2-7B model.
• Minor note: moved float32 conversion in serialize_fp32 from numpy to torch, because numpy does not support BFloat16 dtype (Meta model uses it)
• Let argparse handle one of input model arguments required logic & error message:

usage: export.py [-h] [--version VERSION] (--checkpoint CHECKPOINT | --meta-llama META_LLAMA | --hf HF) filepath
export.py: error: one of the arguments --checkpoint --meta-llama --hf is required

[karpathy]

yay!! 🎉 much nicer.