Add CumulativeROHistogram for fast quantile queries

[thinkingfish]

Summary

Introduces CumulativeROHistogram, a new read-only histogram variant optimized for fast quantile queries using cumulative bucket counts and binary search.

Key Changes

• New CumulativeROHistogram type: A read-only histogram that stores cumulative counts (running prefix sums) instead of individual bucket counts, enabling O(log n) quantile lookups via binary search
• Efficient quantile computation: Uses adaptive search strategy—linear scan for small histograms (≤ cache line size) and binary search for larger ones
• Conversion support: Implements From<&Histogram> and From<&SparseHistogram> for easy construction from existing histogram types
• Rich iteration API: Provides iter() for bucket enumeration and iter_with_quantiles() to get quantile ranges alongside buckets
• Quantile range queries: New bucket_quantile_range() method returns the percentile range each bucket represents
• Comprehensive validation: from_parts() validates index ordering, count monotonicity, and configuration bounds
• Full test coverage: 13 tests covering conversions, quantile matching, edge cases, and validation

Implementation Details

• Stores only non-zero buckets in columnar form (index and cumulative count vectors), matching SparseHistogram's space efficiency
• Implements SampleQuantiles trait for consistent API with other histogram types
• Individual bucket counts are computed on-the-fly from cumulative differences
• Supports serialization via serde and JSON schema generation
• Empty histograms and single-sample cases handled correctly

https://claude.ai/code/session_01TT4hwxFnsHwYthuwR8eTK6