Enable bitsandbytes quantization on AMD GPUs that use warp size 32

[sstamenk]

Purpose

Adds support for bitsandbytes quantized models and Unsloth QLoRA on non-Instinct AMD GPUs that utilize warp size 32.
Requires bitsandbytes #1748 in order to work.

Test Plan

Running models/quantization/test_bitsandbytes.py tests

Test Result

Curently: 7 failed, 1 passed, 4 skipped, 5 warnings

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[mergify]

Documentation preview: https://vllm--27307.org.readthedocs.build/en/27307/

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @sstamenk.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork