sketch_tdigest_vs_materialized_quantile_bench.nx

sketch_tdigest_vs_materialized_quantile_bench.nx -- streaming quantile EXCEED. COMPETITIVE TARGET: Mathematica `Quantile[data, 0.99]`, NumPy `np.quantile(a, 0.99)`, R `quantile(x, 0.99)`, SQL `PERCENTILE_CONT(0.99)`. All require sorted/materialized data. SUBSTRATE: T-Digest delta=100 uses ~32KB regardless of N, and supports DETERMINISTIC MERGE across distributed shards (incumbents cannot merge two quantile arrays meaningfully). HARD-WIN GATE: N=100,000 values streamed. T-Digest p99 within 5% of true. T-Digest memory < 50 KB. Materialized cost: 800,000 bytes (raw values) >= 16x larger.