Add ranged byte[] overloads: compress/decompress(src, offset, length)

[dfa1]

Motivation

Zstd.compress(byte[]) / decompress(byte[]) operate on the whole array. Callers that hold a payload inside a larger buffer (a parsed record, a framing layer) must copy out the sub-range first — wasteful for a binding that prides itself on avoiding copies.

Proposal

Add ranged overloads on Zstd:

public static byte[] compress(byte[] src, int offset, int length);
public static byte[] compress(byte[] src, int offset, int length, int level);
public static byte[] decompress(byte[] compressed, int offset, int length, int maxSize);

Internally these can copy the sub-range into the native input segment directly (no intermediate Arrays.copyOfRange).

Acceptance criteria

• Overloads added with full Markdown /// javadoc (@param/@return) per CLAUDE.md; ./mvnw javadoc:javadoc -pl zstd clean.
• Bounds validated (offset/length against src.length) with a clear exception; covered by tests including corners (offset 0, full length, empty range, out-of-bounds).
• Round-trip + golden/JNI interop tests as appropriate.

Pointers

• zstd/src/main/java/io/github/dfa1/zstd/Zstd.java (existing whole-array methods ~L31-118)

[dfa1]

Won't do — see PR #58 discussion. Ranged byte[] overloads can only ever be a convenience (heap→native copy is mandatory for byte[]), and adding them widens the footgun of using the copying path on hot code. Keeping the byte[] API minimal; zero-copy is the MemorySegment path (and direct ByteBuffer, #56).