Fix node output issues (microsoft#2497)

This pull request updates the `_quantize_matmul` function to improve
compatibility with downstream consumers by preserving the names of graph
outputs during quantization. Now, only intermediate output names are
modified, ensuring that external tools relying on specific output names
continue to work as expected.

Output name handling:

* Updated `_quantize_matmul` in
`olive/passes/onnx/kquant_quantization.py` to only rename intermediate
outputs, preserving the names of graph outputs so downstream consumers
(such as `genai_config.json`) remain compatible.

Author: apsonawane

olive/passes/onnx/kquant_quantization.py

@@ -380,7 +380,10 @@ def _quantize_matmul(
         if accuracy_level > 0:
             kwargs["accuracy_level"] = accuracy_level
 
-        node.outputs[0].name = node.outputs[0].name + f"_Q{bits}"
+        # Only rename intermediate outputs; preserve graph output names so
+        # downstream consumers (e.g. genai_config.json) keep working.
+        if node.outputs[0] not in node.graph.outputs:
+            node.outputs[0].name = node.outputs[0].name + f"_Q{bits}"
 
         return ir.node(
             domain=MSFT_DOMAIN,