feat: downlevel type in index signatures

[davidyuk]

No description provided.

[sandersn]

Please add a Semantics section. Specifically, the transformation as written makes sense for cases where

1. the type is a template literal
2. the type is reference to a template literal
   because x${number}y -> string (for example) is more permissive.

I'm not sure it makes sense for the following types that also work in 4.4:

3. symbol
4. a reference to symbol
5. unions of string, symbol, number or template literals.
6. a reference to unions of string, symbol, number or template literals.

cases (2), (4) and (6) are syntactically ambiguous, so I don't think they can be downlevelled with this tool.
(3) can't be downlevelled at all.
(5) can be downlevelled to multiple index signatures if the symbol isn't in the union.

[davidyuk]

I've updated it to satisfy only the first case.

Regarding making it compatible with the second case, can we add a flag for "unsafe" transformations or allow transformations provided by user?