Fix indentation tokens lengths

[aabounegm]

Removed the duplication of indent tokens as whitespace, and ensured that dedent tokens are 0-width (empty string).
The dedent token could previously contain the matched whitespace of the following indentation (if it matched a previous indentation and didn't introduce a new indent token).

For example, in this snippet:

if True:
    if False:
        print('impossible')
    else:
        print('makes sense')

the dedent token after the first print statement contained the whitespace before the else keyword

Additionally, fixed a minor issue (mostly inconvenience with folding) where the dedent tokens were placed after all newline tokens. For example, in the snippet:

if True:
    print('yes')


else:
    print('no')

the dedent was placed on the 5th line, right before the else keyword. This PR moves it such that it is on the 3rd line.

[msujew]

I think this should not only backtrack through all the newline characters, but also whitespace. For example, if we dedent later:

{
    value
    <-- whitespace
    <-- whitespace
} <-- Dedent token appears at start of this line

This seems pretty weird, as if we remove the whitespace, the dedent already appears in the 3rd line. Since this PR is there to fix behavior such as the folding, I would've thought it should keep the DEDENT token in the 3rd line.

[aabounegm]

But if emit the DEDENT token on the beginning of the 3rd line, then it means that the whitespace after it will be detected as INDENT, and naturally followed by another DEDENT on the next line (and once more, for this example). These extra INDENT/DEDENT tokens are unexpected and will cause parsing errors.

I generally think that trailing whitespace is problematic and should be marked as a parsing error, especially in indentation-sensitive languages.

[msujew]

I see. I guess that makes sense. Thanks for the explanation 👍

[msujew]

Makes sense to me, thanks 👍