Implement UTF-16 based matching

[raskad]

This PR implements UTF-16 based matching to fix #43.

• Add a utf16 feature that enables the UTF-16 api and disables some UTF-8 based optimizations.
• Add a find_from_utf16 api based on the BacktrackExecutor for now.
• Add a Utf16Input InputIndexer.
• Pass the unicode flag to InputIndexer functions, as UTF-16 surrogate handling is based on this flag.
• Move the cursor functions try_slice, try_match_lit and subrange_eq into InputIndexer to avoid InputIndexer functions that have return types which are not directly compatible with UTF-16 input.
• Move some functions out of InputIndexer that where only used internally.

There is a lot more to be done (see https://github.com/ridiculousfish/regress/issues/43#issuecomment-1853268508), but this version allows for UTF-16 support with more 262 tests passing and no regressions in our boa testing. In the next steps I would like to fix the few 262 tests that remain and then add tests for UTF-16 input to regress itself.

[raskad]

Given that unicode is a property of the regex itself, I think this should be attached to the InputIndexer. That is, there should be two different Utf16Input types - one that decodes surrogates, and one that does not. I suggest the names Utf16Input and Ucs2Input respectively. Then we won't need to pass around the unicode flag.

This is a really good idea! I implemented it exactly like that and tested it with boa. Works perfect!