[Issue #1279] Replace C++ metadata serialization with FlatBuffers

[whzruc]

Closes #1279.

This PR supersedes the closed and unmerged #1282. It rebuilds the FlatBuffers migration on the latest master and includes the correctness and integration fixes found during validation.

Summary

• add the shared Pixels metadata schema at flatbuffers/pixels.fbs
• vendor the FlatBuffers toolchain and generate C++ sources during the build
• replace Protobuf metadata handling in the C++ reader, writer, footer cache, statistics, CLI, and DuckDB integration paths
• fix integer decoding, 64-bit RLE handling, asynchronous buffer reads, and CLI error handling uncovered by round-trip testing
• extend writer and CLI test coverage

The existing proto/*.proto definitions remain unchanged for Java and gRPC users. This change only replaces the C++ Pixels file-metadata serialization path.

Validation

• make -j
• IntegerWriterTest: 4/4 passed
• PixelsWriterTest --gtest_also_run_disabled_tests: 7/7 passed
• pixels-cli LOAD generated a FlatBuffers-based .pxl file
• DuckDB pixels_scan successfully read the generated file and returned the expected 3-row aggregate and values

Compatibility note

Files produced by this C++ implementation use FlatBuffers metadata and therefore require the corresponding FlatBuffers-aware C++ reader.

[whzruc]

Null handling was tightened during round-trip validation:

• ColumnWriter now preserves each pixel\u0027s numberOfValues and hasNull fields in the FlatBuffers statistics snapshot.
• IntColumnReader and LongColumnReader expand the compact null bitmap into the output vector and only consume a value for a null row when nullsPadding is enabled. This prevents values after an unpadded null from shifting.

Validated with the 4 integer writer tests, all 7 PixelsWriter tests, and a CLI-to-DuckDB round trip producing 1001, NULL, 1003 for a LONG column.