missing weights input when weights is not a constant input for paddedConv

When testing a model with paddedConv metaop, I get the error: RuntimeError: Missing input#1 for operator Conv2D when compiling the model.
This only happends when input#1 (weights) is not a constant.
Here is the model conv_with_padding.onnx And the test script:

import aidge_core;
import aidge_backend_cpu;
import aidge_onnx
import numpy as np

model = aidge_onnx.load_onnx("conv_with_padding.onnx")

X = np.random.randn(1, 1, 5, 5).astype(np.float32)
W = np.random.randn(1, 1, 3, 3).astype(np.float32)
x_tensor= aidge_core.Tensor(X)
w_tensor= aidge_core.Tensor(W)

model.compile("cpu", aidge_core.dtype.float32, dims=[[1, 1, 5, 5],[1, 1, 3, 3]])
scheduler = aidge_core.SequentialScheduler(model)

# Run inference !
scheduler.forward(data=[x_tensor, w_tensor])
for outNode in model.get_output_nodes():
	output_aidge = np.array(outNode.get_operator().get_output(0))
	print(output_aidge)

changed the description

added Bug 🐛 Important ❗ labels

changed milestone to %aidge v0.7.0

assigned to @hrouis and @pineapple

removed Bug 🐛 label

added Discussion 💬 label

This is currently the expected behavior. Indeed, the compile method tries to set the data type of mandatory parameter inputs. We could relax this constrain, but would it be beneficial for the user?

I changed the status to discussion.

I encountered the issue when adding the metaOp TransposeFC similarly to paddedConv in !350.
In one of the ONNX import tests (gemm_all_attributes) weights and biases are dynamic inputs.
I don't know the risk of relaxing the constraint but it would resolve my test case.

I am in favor of relaxing the error to a Notice or Warning.

I think that the refactor proposed in #268 go in this sens since there will no longer be a difference between "data" and "param" inputs.