Fix various bugs in TypeVarTuples with defaults

[ilevkivskyi]

Fixes a bunch of TODOs in tests.
Fixes #20449

Couple non-trivial things here:

First, I prohibit (regular) type variables with defaults that appear after a TypeVarTuple. The PEP and the spec are clear that these are ambiguous and invalid. But, because they are ambiguous, we can't simply give the error, since otherwise various parts of code will behave in weirdly inconsistent ways. Therefore, I completely erase ambiguous type variables out of existence, 1984 style.

Second, couple commented-out tests cases suggested that TypeVarTuple defaults should be used in cases where *tuple[()] would be used as a value otherwise, plus Foo[X, Y] and Foo[X, Y, *tuple[()]] should be interpreted differently. I don't see anything about this in the PEP/spec, and I don't like this. I think the default should be used only in cases where Any or Never would be used otherwise.

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

static-frame (https://github.com/static-frame/static-frame)
+ static_frame/core/node_re.py:54: error: Unused "type: ignore" comment  [unused-ignore]
+ static_frame/core/node_re.py:59: error: Unused "type: ignore" comment  [unused-ignore]
+ static_frame/core/node_dt.py:120: error: Unused "type: ignore" comment  [unused-ignore]
+ static_frame/core/node_dt.py:124: error: Unused "type: ignore" comment  [unused-ignore]
+ static_frame/core/node_str.py:93: error: Unused "type: ignore" comment  [unused-ignore]
+ static_frame/core/node_str.py:98: error: Unused "type: ignore" comment  [unused-ignore]

[ilevkivskyi]

Not 100% sure what happened with static-frame, but it looks like a clear improvement. Probably caused by the fact that one of the TypeVar values is itself a class generic in a TypeVarTuple with an explicit default.