Reject invalid type ignores

[ilevkivskyi]

Fail on # type: ignore comments we cannot parse instead of interpreting them as "blanket" ignores. This matches old parser behavior (and IMO makes sense). I also fix a bug, where "blanket" type ignore is ignored if there is no space before trailing comment (again this matches old parser and IMO looks consistent).

I only add few tests, as we also have some tests in mypy for this.

[ilevkivskyi]

I found couple more deviations w.r.t. type comments. I am less sure about these two, but they still look quite reasonable:

• In old parser type ignore itself can be ignored with an extra # so e.g. ## type: ignore is not actually a type ignore, in new parser these were still active.
• In old parser, multiple type comments per line is a syntax error, in new one, all except first one were silently ignored.

[ilevkivskyi]

In the interest of speeding things up for the mypy release I am going to merge this later today and update python/mypy#21823. @JukkaL please take a look if you will have a chance.

[JukkaL]

@ilevkivskyi Late review:

• Invalid type ignores like # type: ignore+ seem to behave differently compared to the legacy parser. Can you double check?
• Do cases like x = "s" # type: ignore # type: int behave differently compared to the legacy parser?

[ilevkivskyi]

@JukkaL Thanks! In both cases the behaviour is indeed different, but TBH I am not sure these are actually worth fixing. Let me know if you feel strongly about any of these:

• The error messages are different (invalid type comment in new parser vs invalid type ignore in old parser), but IMO new parser is actually better. The old parser gives the invalid type ignore error for e.g. # type: ignore_, while new one gives name not defined (both parsers give name not defined for e.g. # type: ignored).
• Yes, old parser ignores the type comment, new parser respects it. Again, I am not actually sure it is a bad thing.

[JukkaL]

Yeah, these don't seem worth fixing.