[Fix][ONNX] No precision widening for numpy binary operations

[cbalint13]

This PR enforce some of the ONNX constraint compliance for binary operations.

---

Changes

• Mitigates the behaviour of numpy precision widening, e.g. np.div(int64,in64)->float64
• This complies with the onnx.Div (also others) output constraint: C: Result, has same element type as two inputs
• Adds optional ability to the test checker for dtype similarities, as exact or at least same subtype
• Use a more appropriate public helper for the numpy type conversion, now also working with onnx 0.20

---

Failure example

After shape sampling and arithmetic with division in the graph (excerpt from a custom yolov11s model) np.div(5,2) yields result as float64 (due to numpy behaviour) and then it fails to be a parameter of Slice due to int64 constraint.

tvm.error.InternalError: Check failed: (is_base_of_tuple_of_int64(sinfo)) is false: 
Operator Op(relax.strided_slice) requires the end argument to be a tuple of int64 PrimValues.  
However, in expression {...}, the end argument (R.prim_value(T.float64(2.5)),) has struct info
R.Tuple(R.Prim(value=T.float64(2.5)))

[cbalint13]

Cc @Hzfengsy

[tlopex]

LGTM！