Misc CI/CD updates

[dbarbuzzi]

This PR includes a handful of small fixes for some of the recently updated workflows:

• Include the multi-GPU test job in the WEEKLY run
• Disable a model in the full accuracy test suite (no currently available runners have enough VRAM to run it)
• Increase the server startup wait time (both in accuracy tests and benchmarking)
• Unify the naming for the two accuracy jobs

Including the green remote-push workflow on the latest commit in the branch, there’s also a (mostly) green nightly run here: https://github.com/neuralmagic/nm-vllm/actions/runs/9066138594

• In this run, there appears to be a single test failure which has been failing intermittently (perhaps related to non-determinism of marlin?) unrelated to these changes (tests/models/test_marlin.py::test_models[5-32-half-model_pair0])

[andy-neuma]

cool