This is needed to get proper error reporting in situations where
UnquoteChar is used to decode input stream step by step: there if we see
truncated version of valid character as input always returning ErrSyntax
effectively blocks caller from determining what it was - a real syntax
error or unexpected end of stream.

Unquote error behaviour is preserved to return ErrSyntax always, because
Unquote operates on whole input at once - not as on stream - and
anything wrong there should be seen as really an error in syntax.

Since UnquoteChar is internally used by Unquote, and we already have
Unquote tests to cover all kinds of valid input, we only need to add
tests to cover UnquoteChar error behaviour. Though, unfortunately, we
cannot easily reuse error-behaviour Unquote tests for this.

P.S.

My original use-case for this change is ogórek where UnquoteChar is used
to decode unicode string encoded in Python pickles:

https://github.com/kisielk/og-rek/blob/c7dbf2e4/ogorek.go#L530

Change-Id: I611e7f5795560da488396bc93135a81a56482b75