perf(fsst): close the vortex-jni gap on encode and decode hot paths

[dfa1]

Follow-up hot-path pass over the #287 FSST rewrite, iterated against JavaVsJniFsstBenchmark with per-round profiling (-prof stack:lines=10).

Results (-f 3, same machine/run for both sides)

Benchmark	Before	After	vs vortex-jni
javaFsstEncode	1.848 ops/s	3.059 ± 0.430 ops/s	1.6x slower → parity (jni 2.854 ± 0.289)
javaFsstDecode	27.137 ops/s	32.924 ± 0.824 ops/s	1.3x slower → 1.16x faster (jni 28.353 ± 5.283)

Changes

fsst module (encode hot loop):

• Compressor.compress (both overloads): single unaligned LE 8-byte word load while ≥8 input bytes remain (byteArrayViewVarHandle / LE_LONG); byte-by-byte loadWord retained for the tail only. The fast path drops the over-long-match guard (a ≤8-byte match cannot cross end there) and takes the escape literal from the already-loaded word.
• LossyPerfectHashTable: lookup returns a packed code << 8 | length int (0 = miss) instead of allocating a HashMatch record; table stored as two adjacent longs per slot (symbol + ignoredBits<<16|code<<8|length, the paper's C layout) so a lookup touches one cache line instead of three parallel arrays; empty slots are all-zero and self-answer no-match.
• ShortCodeTable/Matcher: pre-packed no-match sentinel makes the fallback one branch-free array read; longestMatch evaluates both tables and selects (cmov-friendly). codeOf(noMatch) == NO_CODE semantics preserved.

writer adapter:

• FsstEncodingEncoder: all rows compress back-to-back into one shared scratch buffer (was a fresh scratch + exact-size copy per row — 2M allocations per 1M-row chunk write), wire-code remap in a single pass, offsets from rowEnds.

reader adapter:

• FsstEncodingDecoder: output offsets built as per-ptype prefix sums over the uncompressed-lengths child, and the code stream decoded in 256-row batches (per-row ptype switch and two i % cap broadcast modulos removed from the 1M-row loops). Batching is deliberate: a single whole-chunk decompress call measured ~1.8x slower per byte (OSR-compiled mega-loop) — documented on ROWS_PER_DECODE_BATCH.
• New malformed-input guards per the security contract (empty children, invalid code offsets, row-count and decoded-length overflow, decoded-vs-claimed length mismatch), each with a negative test in the new FsstEncodingDecoderTest (10 tests).

Verification

• ./mvnw verify green (all modules, integration tests included) — Rust interop (JavaWritesRustReads, RustWritesJavaReads) confirms wire output unchanged.
• vortex-reviewer pass: APPROVE; its should-fix (negative tests for the new guards) and nit (>2GB decoded-length guard) are included here.
• Docs: CHANGELOG entry with the -f 3 table; TODO FSST follow-ups refreshed (stale 1.6x-gap phrasing).

🤖 Generated with Claude Code

[dfa1]

Second review pass — one bug found and fixed

Re-reviewed the full diff line-by-line. Verified the algorithm changes are byte-preserving (compressor fast path, packed hash slots, ShortCodeTable/Matcher, encoder shared-scratch — all equivalent to the prior code, confirmed against the Rust-interop tests). Found and fixed one real defect the first pass missed:

FsstEncodingDecoder: out-of-bounds read on a constant-encoded code-offsets child. The codesOffCap == 1 (constant-broadcast offsets) case was bundled into the batched fast path, whose loop reads readUnsigned(codesOffsetsSeg, batchEndRow) at row indices 256, 512, … — with no modulo. For a column of >256 empty strings (all offsets equal → offsets child collapses to one physical element), that runs off the 1-element segment and throws a raw IndexOutOfBoundsException, violating the security contract (must be VortexException). The pre-rewrite code was safe here via i % codesOffCap; the batched rewrite dropped the modulo.

Fix: cap == 1 now has its own branch — every row range is empty [off, off), so the column decodes to nothing and a non-zero claimed length is rejected as malformed. Added two regression tests (constantCodeOffsetsManyRows_decodesEmptyColumn at n=300, and the malformed-length counterpart). FsstEncodingDecoderTest now 12 tests; ./mvnw test -pl fsst,reader,writer -am green.