Actually make the run_compressed test useful

[dsikka]

SUMMARY:

• Make the test case useful
• Previously testing identical model loads as run_compressed can't be turned off through the HFQuantizer pathway. Use ModelCompressor in its place until this is supported
• Update prompt to be sensical

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

[horheynm]

Nice Ill fix it up once the decompress pathway is solid on HFQuantizer

[horheynm]

Note that
we need #923
to load the model using run_compressed=False
and
HF side huggingface/transformers#34719
for setting status to frozen

to address the TODOs on the PR