GH-1116: [Java] Fix compressed buffer prefix write and ZSTD dstCapacity

[LuciferYang]

What

Two fixes in the compression codec:

1. AbstractCompressionCodec.compress(): uncompressed-length prefix could be written with a wrong value

   The old code read uncompressedBuffer.writerIndex() after doCompress() to populate the 8-byte prefix. But uncompressedBuffer is a shared reference to the vector's internal buffer. If writerIndex changed between doCompress() and the subsequent read -- e.g. due to VectorSchemaRoot reuse (clear/allocateNew) -- the prefix would get a wrong value such as 0.

   Fix: capture writerIndex() once at the top of compress() and reuse it for the empty-buffer check, size comparison, and prefix write.

2. ZstdCompressionCodec.doCompress(): dstCapacity overstated by 8 bytes

   Zstd.compressUnsafe(dst, dstSize, ...) expects dstSize to be the available space from dst. The code offsets dst by 8 bytes past the prefix, but passed 8 + maxSize instead of maxSize. In practice compressBound() headroom prevented an actual out-of-bounds write, but the parameter was semantically wrong.

   Fix: pass maxSize instead of dstSize.

Tests

• testMultiBatchZstdStreamWithWideSchemaAndAllNulls -- 100 fields x 10 batches x 500 rows, VectorSchemaRoot reuse with all-null timestamp columns in every 3rd batch, full streaming round-trip with per-cell verification.
• testAllNullFixedWidthVectorZstdRoundTrip -- 3469-row all-null TimestampMilliVector, buffer-level compress/decompress, asserts decompressed writerIndex matches the original.

Closes

Closes #1116

[github-actions]

Thank you for opening a pull request!

Please label the PR with one or more of:

• bug-fix
• chore
• dependencies
• documentation
• enhancement

Also, add the 'breaking-change' label if appropriate.

See CONTRIBUTING.md for details.