convert to double before float materialization in scalarize shapes

[zjgarvey]

Addresses a bug when trying to materialize a non fp64 attr to a constant float op in scalarize shapes.

[jinchen62]

@zjgarvey I'm a bit confused, why are we converting all FloatAttr to double.

[zjgarvey]

@zjgarvey I'm a bit confused, why are we converting all FloatAttr to double.

Torch::ConstantFloatOp has a value attribute that needs to be a f64 or double. Not converting FloatAttrs to double and trying to create a constant float op will result in the error you saw:

error: 'torch.constant.float' op attribute 'value' failed to satisfy constraint: 64-bit float attribute
    %908 = torch.operator "onnx.Cast"(%907) {torch.onnx.to = 1 : si64} : (!torch.vtensor<[1],si64>) -> !torch.vtensor<[1],f32>
note: see current operation: %777 = "torch.constant.float"() <{value = 3.000000e+00 : f32}> : () -> !torch.float

[jinchen62]

LGTM