Fix aten_linear: squeeze trailing dim after 1D-weight MatMul

[Copilot]

When weight is 1D, aten_linear unsqueezed it to (in_features, 1) for MatMul but never squeezed the resulting trailing dim back off. This produced output rank input.rank instead of the correct input.rank - 1, mismatching PyTorch eager's dot-product contraction semantics.

Changes

• onnxscript/function_libs/torch_lib/ops/nn.py: Apply op.Squeeze(..., [-1]) immediately after the 1D-weight MatMul, before the bias is None branch, so both paths return the correctly-ranked result:

if len(weight.shape) == 1:
    weight_transposed = op.Unsqueeze(weight, [1])
    mul = op.Squeeze(op.MatMul(input, weight_transposed), [-1])
else:
    weight_transposed = op.Transpose(weight, perm=[1, 0])
    mul = op.MatMul(input, weight_transposed)
if bias is None:
    return mul
return op.Add(mul, bias)

• tests/function_libs/torch_lib/extra_opinfo.py / ops_test_data.py: Add nn.functional.linear_1d_weight OpInfo with sample_inputs_linear_1d_weight, covering 1D/2D/3D inputs with and without bias. The 2D-input+bias combination is intentionally excluded — PyTorch eager itself rejects it.

• Fixes aten_linear doesn't squeeze back the dim added for 1D-weight MatMul when bias is None #2982