Support output QK symbolic shape inference

[derdeljan-msft]

Update GQA symbolic shape inference (

onnxruntime/onnxruntime/python/tools/symbolic_shape_infer.py

Line 2548 in 9de58ac

def _infer_GroupQueryAttention(self, node): # noqa: N802

) to account for newly added output QK optional output.