Fix inconsistent IR when printing R.nn.pad operator

[parsifal-47]

Hi All,
This is a fix for #17483
the inconsistency caused by pad_width and pad_value being second argument, this PR moves pad_value to the attributes.
Let me know if that is not a proper way to resolve the issue, additionally I am not sure if there is a way to make a test for show, I would appreciate references in this case.

Thank you!

[yongwww]

@tvm-bot rerun

[yongwww]

@tvm-bot rerun

[parsifal-47]

@tvm-bot rerun

[Cookiee235]

@tvm-bot rerun

[parsifal-47]

@Cookiee235 I do not understand the workflow completely, but it looks like some approval is needed in order to run the rest of the PR checks

[yongwww]

@tvm-bot rerun

[parsifal-47]

tests are working now, but when going from tensorflow in tests/python/contrib/test_msc/test_translate_tensorflow.py the constant is not getting folded and remains Var or getting converted to Var for some reason, not sure how to fix that.

I tried adding transforms to AttrCvt with fold_constant, but it looks like the Var appears somewhere after.

[parsifal-47]

to elaborate a bit more, originally, when pad_value was passed as positional argument, it was folded automatically, but because pad_width was actually the second positional argument, printer was broken. After I moved it to attributes, automatic folding no longer applies and it looks like legalizer is not a good place to do that because I do not see any other foldings there. Looks like the best place is in conversion from tf because other tests are green, but I tired multiple combinations and haven't found any workable, will take another look in a few days, meanwhile if you have an advice, let me know, thank you!