Stream index creation through byte arrays

[platypii]

Summary

• consume Hyparquet column chunks directly instead of materializing row objects
• preserve UTF-8 source values as byte arrays during the numeric-token indexing path
• extract tokens directly from UTF-8 and write shared token byte arrays without decode/re-encode cycles
• require hyparquet-writer 0.16.4 for byte-array page statistics
• report peak RSS in the WildChat build benchmark

WildChat benchmark

• before byte-array output: 37m48.8s, 21.14 GB peak RSS
• after: 28m29.2s, 17.16 GB peak RSS
• improvement: 24.7% less time and 18.8% less peak memory
• output is byte-for-byte identical to the prior index

The previous JavaScript baseline was 36 minutes; this build completes about 20.9% faster.

Testing

• npm test (155 tests)
• npm run lint
• npx tsc --noEmit
• git diff --check