Support round-tripping non trivial glue

[mcculls]

What Does This Do

Our current defineClass glue is safe to unpack using getBytes(UTF_16BE) because it has no unpaired surrogates.

However we should make sure the general Glue.unpackBytecode helper can tolerate non-trivial glue that may contain unpaired surrogates. To do this we unpack the characters into bytes ourselves, avoiding the JDK's built-in surrogate validation.

Contributor Checklist

Jira ticket: [PROJ-IDENT]

[ygree]

I noticed a couple of issues with the current glue bytecode packing scheme: it treats arbitrary byte pairs as UTF-16BE code units, which means the generated bytecode has to be padded to an even length and decoding can run into invalid/surrogate UTF-16 values unless handled carefully. That makes the implementation more subtle than it needs to be.

I spiked an alternative Latin-1 implementation in #124 where each byte maps directly to one character and decoding is just String.getBytes(ISO_8859_1). This removes the padding requirement and avoids the pseudo-UTF-16BE/custom-decoding concern.

I approve, and I'll leave it up to you to decide which approach to use.

[mcculls]

@ygree yes, your approach does make more sense for the glue we're generating, so I'm going to drop my PR.

Originally I was concerned that we could end up with very large string allocations, but it turned out we only need to use this approach for a limited number of classes. And even the extra ones I'll be adding soon are not large enough that using ISO_8859_1 would be that noticeable, as you point out.

One ask - currently I can't merge your PR because the commit is missing a verified signature - can you fix that?