const evaluator: string values and init operations

[marcrasi]

Adds string values, string initialization operations (using a "WellKnownFunction" mechanism), and some tests that invoke string initialization but do not check the contents of the strings.

This very basic string handling is all that we have in the tensorflow branch const evaluator right now.

[ravikandhadai]

@marcrasi I wonder if it would be better to make a separate function for lines 560 to 591 that handles well-known functions. As we plan to add more well-known functions, it would make this more modular.

[marcrasi]

done!

[ravikandhadai]

@marcrasi LGTM otherwise.

[ravikandhadai]

Practically speaking, though, would it be better to represent those common collection types in a more symbolic fashion for compile-time evaluation? Especially if the compile-time evaluator has a cap on number of operations or some other complexity cap, interpreting the standard library implementations at such a low level seems like it could easily blow that cap, or else just be painfully slow to interpret in practice.

@jckarter Related to your suggestion on handling collection types symbolically, I just wanted to give heads up that this PR adds basic support for symbolically representing strings and interpreting their initializers. Representation for arrays and array initialization is added by this PR #21703

[ravikandhadai]

@swift-ci Please smoke test

[devincoughlin]

LGTM.

[ravikandhadai]

@swift-ci Please test and merge

[ravikandhadai]

@swift-ci Please test and merge

[ravikandhadai]

@marcrasi Once this is merged, can we upstream enums before merging arrays?