feat(inmemory): add Rayforce grouped and parted adapter

[singaraiona]

Summary

• add Rayforce as an in-memory NYSE TAQ query engine
• import the full trade and quote schemas into a native Rayforce splayed database
• implement all 53 benchmark queries
• benchmark both grouped (time) and parted (sym,time) layouts
• support query filtering, persisted correctness output, and standard merged PSV results

Rayforce dependency

This adapter requires the fixes merged in RayforceDB/rayforce#326. Build Rayforce from clean release objects:

cd /path/to/rayforce
make clean
make release

Set RAYFORCE_BIN when the Rayforce and benchmark repositories are not sibling directories:

export RAYFORCE_BIN=/absolute/path/to/rayforce/rayforce

Data preparation

Rayforce needs the new full-schema database; databases produced by the earlier reduced-schema adapter must be regenerated. Generate the KDB database first because the Rayforce converter exports its source data from the sibling kdb directory:

DATAFORMAT=kdb ./generateDB.sh \
  "${NYSEBENCHMARKDIR}/${SIZE}/psv" \
  "${NYSEBENCHMARKDIR}/${SIZE}/kdb" \
  "${DATE}"

DATAFORMAT=rayforce RAYFORCE_BIN=/absolute/path/to/rayforce/rayforce \
  ./generateDB.sh \
  "${NYSEBENCHMARKDIR}/${SIZE}/psv" \
  "${NYSEBENCHMARKDIR}/${SIZE}/rayforce" \
  "${DATE}"

Run Rayforce alone with:

./benchmarks/inmemory/queryEngines.sh \
  --db-dir "${NYSEBENCHMARKDIR}/${SIZE}" \
  --param-dir "./artifacts/parameters/${SIZE}" \
  --date "${DATE}" \
  --threads "0 4 16 64" \
  --engines rayforce

Validation

• grouped output matches KDB-X for 53/53 queries
• parted output matches the verified parted reference for 53/53 queries
• grouped, 0-secondary-thread run: KDB-X 3.584 s, Rayforce 2.880 s (1.245x total speedup; 27/53 query wins)
• parted, 0-secondary-thread run: KDB-X 2.931 s, Rayforce 2.490 s (1.177x total speedup; 33/53 query wins)
• post-upstream-sync smoke test passed for grouped and parted layouts
• shell syntax and whitespace checks pass

The timing figures are from the small full-schema validation dataset and are intended as a regression/performance sanity check, not as final published benchmark numbers.