Fix autoawq build failure by pinning to version with wheels

The nightly pipeline was failing when pip tried to build autoawq==0.2.8
from source, which requires torch to be available during the build process.
With torch==2.8.0, pip's dependency resolver was selecting autoawq==0.2.8
which needs to build from source.

Pinning to autoawq==0.2.9 ensures pip uses the pre-built wheel, avoiding
the build-time torch dependency issue. This is cleaner than using
--no-build-isolation and maintains proper build isolation.

Note: autoawq is used in partition.py for AWQ quantization.

Fixes: ModuleNotFoundError: No module named 'torch' during autoawq build

Author: redhairdragon

serving/docker/lmi-container-requirements.txt

@@ -23,7 +23,7 @@ scipy==1.16.0
 onnx==1.19.0
 sentence_transformers
 onnxruntime-gpu==1.20.0
-autoawq
+autoawq==0.2.9
 tokenizers
 pydantic==2.11.7
 optimum==1.23.2