Use ISO-8859-1 for glue bytecode packing

[ygree]

What Does This Do

This removes the padding requirement, avoids the pseudo-UTF-16BE/custom-decoding concern, and slightly reduces the compiled runtime footprint, even though the generated Java source gets larger.

Motivation

ISO-8859-1 gives a direct byte-to-char mapping, so glue decoding becomes a simple packed.getBytes(ISO_8859_1) with no even-length padding requirement and no surrogate edge cases.

Additional Notes

The source gets larger:
Raw generated DefineClassGlue.java is larger: 12,278 B -> 17,268 B.

Everything that is compiled or packaged got smaller:
ClassInjector$MethodPatch.class: 8,031 B -> 7,749 B
class-inject runtime jar: 8,836 B -> 7,894 B
utils jar: 31,030 B -> 30,914 B
even compressed class-inject sources jar got smaller: 7,007 B -> 6,248 B, because octal escapes compress well.

One caveat: on Java 8, the in-memory String object for the packed glue may be larger with Latin-1 because Java 8 stores strings as char[], so twice as many chars means roughly twice the string backing storage. On Java 9+ compact strings likely make it similar or smaller. But this is a tiny, transient-ish class-injection path, while the shipped class/jar footprint is smaller.

Contributor Checklist

Jira ticket: [PROJ-IDENT]

[mcculls]

Good suggestion, for the glue we're generating this does indeed make more sense