feat: Migrate block compression infrastructure with LZ4, ZSTD, and none backends

[lxy-9602]

Purpose

No Linked issue.

Migrate block compression framework for SST file I/O:

Block compression core (src/paimon/common/compression/):

• BlockCompressionType — compression type enum (LZ4, ZSTD, NONE) (block_compression_type.h)
• BlockCompressor — abstract compressor interface (block_compressor.h/cpp)
• BlockDecompressor — abstract decompressor interface (block_decompressor.h/cpp)
• BlockCompressionFactory — factory for creating compressor/decompressor by type (block_compression_factory.h/cpp)
• NoneBlockCompressionFactory — passthrough (no-op) compression factory (none_block_compression_factory.h)

LZ4 backend (src/paimon/common/compression/lz4/):

• Lz4BlockCompressionFactory — LZ4 compression factory (lz4_block_compression_factory.h)
• Lz4BlockCompressor — LZ4 compressor implementation (lz4_block_compressor.h)
• Lz4BlockDecompressor — LZ4 decompressor implementation (lz4_block_decompressor.h)

ZSTD backend (src/paimon/common/compression/zstd/):

• ZstdBlockCompressionFactory — ZSTD compression factory (zstd_block_compression_factory.h)
• ZstdBlockCompressor — ZSTD compressor implementation (zstd_block_compressor.h)
• ZstdBlockDecompressor — ZSTD decompressor implementation (zstd_block_decompressor.h)

Tests

• block_compression_factory_test.cpp — factory creation, round-trip compress/decompress for LZ4, ZSTD, and NONE

API and Format

Documentation

Generative AI tooling

Migrate-by: Aone Copilot (Claude)

[lxy-9602]

Thank you @ChaomingZhangCN for the contributions to the block compression infrastructure — including the core framework, LZ4/ZSTD/NONE backends, and tests — migrated as part of this batch. 🎉

[leaves12138]

I found a blocking issue in the LZ4 block decompressor: it reads the 8-byte header before validating that src_length is at least HEADER_LENGTH. If a truncated/corrupted block shorter than 8 bytes is passed in, Decompress will read past the supplied buffer instead of returning an Invalid status. Please add a src_length >= HEADER_LENGTH guard before ReadIntLE and cover it with a corruption test. No other blocking issue found in this pass.

[leaves12138]

Re-reviewed the latest revision. The previous blocking issue is fixed: the LZ4 decompressor now validates that the input is at least the header length before reading the header, and the added regression coverage exercises the truncated-header and insufficient-output-buffer cases. I did not find any remaining blocking issue within the staged migration scope. Approving.