Hard release with supporting scripts and working harness

1. We have the harness supporting both the Offline and the Server for gpt-oss-120b
2. We would need to update LoadGenServerClient to handle race conditions and timing issues .
3. We are observing negative TPOT and TTFT primarily due to the race condition between the worker thread sending first token and the other thread which sends the response before .
4. This release containing submission automation scripts both to run submission and create correct submission structure .

Script to launch automated runs

Scripts written to launch and convert into submission edible format

Compliance and accuracy fixes Offline/Server

This is an important release which fixes compliance failures observed

1. GPT-OSS-120B has two compliance tests TEST07 and TEST09 .
2. TEST07 verifies accuracy during performance run .
3. The returned text from vllm was detokenized and sent as response missing out on special tokens .
4. For Server streaming case we were also sending the First Token twice
5. Using "return_token_ids": True , got the token_ids and sent them as a loadgen response .
6. This fixes the accuracy issues seen so far .

Fix offline async concurrency issue

This recent fix helps to fix offline async concurrency hang . Where the client used to hang for a higher thread count . The issue is to cleanly separate the threading aspect and the allow the main thread to update to loadgen

basic functionality working

Working SUT . Need few more changes

Added concurrency specifier

Allow the harness to specify offline-async-concurrency . Uses a big of threads to issue individaul http requests.

Accuracy Check for gpt-oss-120b

This release has a working accuracy check for offline scenario .