Add SmileGenerator.Feature.LENIENT_UTF_ENCODING for lenient handling of broken Unicode surrogate pairs on writing

[kireet]

When encoding some invalid user generated data, we encounter JsonGenerationExceptions with the message Unmatched first part of surrogate pair. This didn't occur when using text serialization.

It seems there's a CBOR option to avoid this exception, CBORGenerator.Feature.LENIENT_UTF_ENCODING, but not one for Smile. It would be great to add this feature.

To reproduce, just try to serialize an invalid string, such as "\uD83D".

[cowtowncoder]

Sounds like a good idea, hoping to implement when I have time!

Note: was done for CBOR via #222.