Perf: replace O(n) linear scans with O(log n) binary search across sorted element collections.

[mohammedahmed18]

Profiling Findings

CPU profiling of the UpdateElements (heaviest write path) revealed:

• runtime.memmove at 16.2% from slice copies in Insert/Delete
• Elements.Insert at 14.5% flat / 27.5% cumulative -- O(n) linear scan
• Elements.Delete at 11.5% flat / 20.5% cumulative -- O(n) linear scan
• Elements.FindIdx at 7.2% -- O(n) linear scan for element ID lookup
• GC pressure at 13.8% from frequent allocations

Optimizations

elements.go:

1. Binary search in Elements.Insert -- O(n) to O(log n) for finding insertion point
2. Binary search in Elements.FindScoreIdx -- O(n) to O(log n) for score lookup
3. Binary search in Elements.FirstElementsWithScore -- O(n) to O(log n) per score
4. Eliminated redundant scans in Elements.Add -- was: FindIdx (O(n)) + Delete (O(n)) + Insert (O(n)); now: single scan + binary insert
5. Eliminated redundant scans in Elements.Update -- was: Delete (O(n)) + Insert (O(n)); now: combined delete + binary insert

segment.go:
6. Binary search in Segment.FilterIdx -- O(n) to O(log n) for finding element range within segment
7. Binary search in Segment.FilterScoresIdx -- O(n) to O(log n) for score range

Benchstat Results (statistically significant, p < 0.05)

Benchmark	Baseline	Optimized	Improvement
UpdateElements/1K	405.8us	332.0us	-18.18%
UpdateElements/10K	9.186ms	7.204ms	-21.58%
GetElements/1K	114.0us	112.2us	-1.54%
DeleteElements	2.267ms	2.211ms	-2.49%
BulkIngestion/10K	7.071ms	7.017ms	-0.78%
Geomean (all)	803.8us	770.5us	-4.15%

[mohammedahmed18]

please take another look @volodimyr

[glaslos]

@mohammedahmed18 do you have benchmarks you could add to the repo?

[mohammedahmed18]

@glaslos added the benchmarking file, let me know if there are other changes you want me to make