[DRAFT] LLM Compressor integration

[idoudali]

Description

Related issues

Fixes # (issue)

Before submitting

• This PR only changes documentation. (You can ignore the following checks in that case)
• Did you read the contributor guideline Pull Request guidelines?
• Did you link the issue(s) related to this PR in the section above?
• Did you add / update tests where needed?

Reviewers

At least one review from a member of oumi-ai/oumi-staff is required.

---

Summary by Gitar

• Major integration change:
  • Replaced AWQ quantization backend with LLM Compressor (vLLM project), supporting FP8, GPTQ, AWQ, and W-bit schemes
  • Deprecated awq_quantizer.py; added llmcompressor_quantizer.py with oneshot API integration
• Configuration overhaul:
  • Changed from method (string) to backend + scheme + algorithm (enums) in QuantizationConfig
  • Added calibration dataset control, ignore-layers, and algorithm selection
• CLI updates:
  • Split --method into --backend, --scheme, --algorithm flags
  • Updated example model IDs and quantization examples
• Tests:
  • Removed AWQ tests; added LLM Compressor and constants tests
  • Updated builder, BNB, and base quantization tests for new enum-based API

_{This will update automatically on new commits.}

[idoudali]

we would like not to expose the backend

The customer cares more about the format and not the backend being use

Keep the naming convention as is for now and infert the backend form the format. use a prefix like "bnb_" for the old case to avoid introducing more changes