Bit-packing encode costs ~2x Rust's CPU #475

Bit-packing is the largest remaining encode-CPU gap against Rust: ~277 ms vs ~142 ms per 10M-row OHLC cascading write (async-profiler, measured while working on #464–#469).

• BitpackedEncodingEncoder.encode scans every value for min/max and a bit-width histogram (split into two loops in perf(writer): align cascade stats with Rust's compressor (dense/run-aware counts, lazy ArrayAndStats) #464). Rust computes only the histogram here (bit_width_histogram, ~103 ms) and takes min/max from the array's cached stats.
• packFastLanes is a scalar per-lane loop with a typeBits switch on every word read/write (readWordFromSeg / writeWordToSeg).

Levers: pass min/max from the cascade's ArrayAndStats (the #464 hook) instead of rescanning; specialize packFastLanes per typeBits and bitWidth (as the decoder side does). Output must stay byte-identical.