onnx model constrains input shape

[FynnBe]

Also the input size for the onnx model seems to be constrained to [1, 1, 256, 256] only

Originally posted by @carlosuc3m in bioimage-io/collection-bioimage-io#540 (comment)

It seems like we need to allow onnx weights to specify/overwrite their own input shape(s).

something like this probably:

...
weights:
    onnx:
        ...
        inputs: [{shape: [1, 1, 256, 256}]

[FynnBe]

alternatively (and ideally) we figure out how to enforce that the batch dimension has shape "-1" (which I think is the onnx convention to indicate an arbitrary size).
In addition it would be nice to find a convenient way to convert existing, fixed-batch-size onnx weights.