Use PIL Image internally for the Multimodal Agent (#1124)

* Change defualt model for `lmm`

* Try to use PIL image for LMM's _oai_messages

* Update test cases and llava

* Remove redundant files

* Update the imports for lmm tests

* Test case fix

* Docstring update

* LMM notebook lint

* Typo correction for img_utils and its test

* Update test_llava.py

debug, reformat

---------

Co-authored-by: Chi Wang <wang.chi@microsoft.com>
Co-authored-by: Shaokun Zhang <shaokunzhang529@gmail.com>
Co-authored-by: Shaokun Zhang <shaokun.zhang@psu.edu>