fix: only raise TC101 for str literals

[shiftinv]

This PR changes the code related to the TC101 check to only show errors if there's a str literal/constant, not any other type.

There are valid use-cases for literals that don't represent types in annotation subscripts, especially with the introduction of typing.Annotated, like x: Annotated[int, 42] or generally custom types with __class_getitem__.

Adding on to this, I suppose Annotated[int, "stuff"] should technically also be allowed, but it's not possible to generally differentiate between contexts where a string literal is meant to represent a type (e.g. Dict[str, "int"]), and where it doesn't (e.g. Annotated[str, "int"]), without hardcoding specific generic types.
As such, this is by far not a perfect solution, but it should get a bit closer :)

For context, this first appeared here:
https://github.com/DisnakeDev/disnake/blob/dd1fb9e4cbfd5974bf4e2ddf001aecc7b6443478/examples/slash_commands/param.py#L76-L78

./examples/slash_commands/param.py:76:29: TC101 Annotation '1' does not need to be a string literal
./examples/slash_commands/param.py:76:32: TC101 Annotation '10' does not need to be a string literal
./examples/slash_commands/param.py:77:35: TC101 Annotation '0' does not need to be a string literal
./examples/slash_commands/param.py:78:30: TC101 Annotation '0' does not need to be a string literal
./examples/slash_commands/param.py:78:33: TC101 Annotation '1.0' does not need to be a string literal

[codecov]

Codecov Report

Merging #122 (12148a9) into main (7be3641) will not change coverage.
The diff coverage is 100.0%.

@@          Coverage Diff          @@
##            main    #122   +/-   ##
=====================================
  Coverage   98.7%   98.7%           
=====================================
  Files          3       3           
  Lines        466     466           
=====================================
  Hits         460     460           
  Misses         6       6           

Impacted Files	Coverage Δ
flake8_type_checking/checker.py	98.5% <100.0%> (ø)

[sondrelg]

I think your example for where it doesn't might be wrong, but PR looks good! I can't think of a reason why we wouldn't limit this to string literals.

And wrt. hardcoding generic types. It seems like it could make sense to add a set of hardcoded common types at some point, for this purpose.