Decode \u surrogate pairs in quoted strings

[chiliec]

Fixes #83

readString decoded each \uXXXX escape as its own rune, so a UTF-16 surrogate pair ("\ud83d\ude00", the way JSON encoders write non-BMP characters) came out as two U+FFFD. Now, when a \u escape is a high surrogate and is immediately followed by a \u escape for a matching low surrogate, the two are combined with utf16.DecodeRune, which matches encoding/json. Lone or mismatched surrogates still decode to U+FFFD like they do today, and a bad hex digit after \u still returns the existing "Bad \u char" error. The hex-digit parsing moved into a small hexValue helper so it can be reused for the lookahead.

Tests: added TestUnicodeSurrogatePair, which covers a pair (lower and upper case), a pair inside an object value, and lone/reversed/unpaired surrogates.

go test -run TestUnicodeSurrogatePair .   # before the fix: FAIL (3 pair cases give "\ufffd\ufffd")
go test ./...                             # with the fix: ok

[trobro]

Thanks for the PR! I moved test to the asset files, they are easier to exchange between different implementations of Hjson. I removed tests that resulted in \ufffd output, that's not really a desired output, could be considered a bug.

[chiliec]

Thanks @trobro — the shared asset files are clearly the better home for these.

On the \ufffd cases: those are lone/unpaired surrogates, and I mapped them to U+FFFD on purpose since that's what Go's own unicode/utf8 and encoding/json do with invalid surrogates. Fine to leave out of the shared suite until Hjson settles the canonical behavior — happy to switch it to error on a lone surrogate instead if you'd prefer.

[trobro]

The behavior of encoding/json differs between v1 and v2: v2 rejects inputs with invalid UTF-8. But I don't want to do that breaking change in Hjson.

[chiliec]

Makes sense. Keeping v1's U+FFFD behaviour avoids breaking existing inputs. Thanks for merging!