Add support for Int128 in codegen and macros

[BlobCodes]

This PR adds:

• Ability to use U/Int128 in macros
• MathInterpreter U/Int128 support (for ENUMs and compile-time interpreted generics)
• U/Int128 LLVM inlining (probably improves performance?)

Things that still need to be addressed:

• The macro to_i method always converts a string to a int64, and then casts it to a int32. integers greater than Int64::MAX return an error "(number) is not an integer", while integers greater than Int32::MAX but smaller than or equal Int64::MAX return an error "Invalid Int32: (number).
• Crystal::Call#convert_numeric_argument does not seem to support int128, but I don't know what it is used for.

Probably requires #11551 to pass compiler specs on windows.

[watzon]

Seems like the only failing specs currently are in the interpreter due to a missing Syscall::GRND_NONBLOCK. Does this still seem related?