In-house FlatBuffers toolchain: drop flatc + com.google.flatbuffers (ADR 0017)

[dfa1]

Replaces the external flatc compiler and the com.google.flatbuffers runtime with an in-house, MemorySegment-native FlatBuffers toolchain — mirroring the existing proto-gen. Implements ADR 0017.

Why

1. MemorySegment-native end to end. The reader memory-maps the file into one MemorySegment; the generated readers/builder now operate on it directly — no ByteBuffer bridge in the codec, zero-copy slices.
2. Unblocks JPMS / clean Maven Central. flatbuffers-java shipped as an automatic module (filename-derived name), which a published named module cannot reliably requires. With both wire formats self-hosted, no upstream automatic module remains → module-info.java is now viable.

Also drops the external flatc CLI (no brew install, no ValidateVersion version-guard strip hack) and restores the four .fbs schemas to upstream-pristine names (the Fbs prefix now lives in the generator).

What

• fbs-gen module — .fbs lexer/parser/AST + code generator (mirrors proto-gen).
• io.github.dfa1.vortex.fbsrt runtime — FbsTable / FbsStruct (read) + FbsBuilder (write) over MemorySegment. Full FlatBuffers algorithm: vtable lookup/dedup, back-to-front build, alignment, default omission, unions, inline structs, strings.
• Generated classes carry @Generated; output names unchanged (FbsArray, …), so no consumer churn beyond the API-shape moves (getRootAsX(MemorySegment), byte-vector xAsSegment()).
• regenerate-sources profile runs fbs-gen (+ proto-gen) execs; flatc exec + antrun strip removed.
• Cleanups: killed MemorySegment→ByteBuffer→MemorySegment double-conversions in the flat-segment decoders; migrated Layout.metadata to MemorySegment.

Validation

• Byte-identical to flatc (builder validated against the flatc builder for the same sequences).
• Full unit suite + 275 integration tests (Rust interop: Rust reads our output, we read Rust's) pass.

Follow-ups (not in this PR)

• Add module-info.java (now unblocked) and non-export the fbs/proto packages.
• Finish the ByteBuffer purge: decoder ctx.metadata() (~25 decoders) and ZonedStatsSchema.

🤖 Generated with Claude Code