Symbols need to support UTF-8 names

MRI currently supports UTF-8 symbols:

```
x = :"\u{3042}" # => :あ
```

However, JRuby always treats symbols as US-ASCII, resulting in things like encoding errors when trying to inspect one:

```
> x = :"\u{3042}"
ArgumentError: invalid byte sequence in US-ASCII
from org/jruby/RubySymbol.java:221:in `inspect'
```

This is a source of a number of errors in the new stdlib test suite.
