perf: iterate counts[] in value_at_quantile to elide bounds checks

[fcostaoliveira]

Summary

The value_at_quantile prefix-sum scan iterated with for i in 0..self.counts.len() and indexed
self.counts[i]. Iterating the slice directly with self.counts.iter().enumerate() hands the
element to the loop without a per-element bounds check on the hot path. Results are unchanged.

Benchmark

gnr1 (Intel Granite Rapids), single core, same-session A/B:

	before	after	Δ
value_at_percentile (Mq/s)	0.1742	0.1830	+5.1%

Percentile results byte-identical (the benchmark sink is unchanged). cargo test green;
cargo fmt --check clean.

[fcostaoliveira]

Superseded by #140. That PR rewrites the same value_at_quantile scan as a chunked skip-scan (sum a chunk, skip if it can't reach the target, walk only the crossing chunk), which iterates chunks/elements and therefore already elides the per-element bounds check this PR targeted — a strict superset. It measures +63% read (vs the ~+5% here) with a byte-identical parity test across all counter widths. Closing in favor of #140.