Add Logging Utils

[mvanniasingheTT]

Adding Logging

This PR adds logging functionality to the online and offline mock models as well as to the vllm api server. Note there was a previous PR with the same changes but this PR is rebased off main and has a cleaner commit history.

Note:

There is some repeated code from logging_utils.py in mock_vllm_model.py since usage of these functions and classes in the online and offline mock model inference scripts and requires refactoring of the tt-inference-server directory naming. Repeated code can be removed following the refactoring.

Changes

This PR addresses comments from the previous PR such as:

1. Handling the case where only decode is happening
2. Using JSONL vs JSON for less overhead
3. Cleaner patches
4. Handles the case where not all requests come in at same time for online inference