Add exponential and vectorized search functions from Interpolations.jl

[ChrisRackauckas-Claude]

Summary

This PR addresses issue #8 by implementing search functions inspired by Interpolations.jl's specialized index finding routines.

New Functions Added:

• searchsortedfirstexp(v, x, lo, hi): An exponential search followed by binary search, optimized for correlated sequential lookups where the target is expected to be near the starting position lo. This is similar to Interpolations.jl's searchsortedfirst_exp_left.

• searchsortedfirstvec(v, x): Batch search for multiple sorted values. When the input vector x is sorted, this leverages monotonicity to avoid redundant searching by reusing the previous search result as the starting point.

• searchsortedlastvec(v, x): Similar to searchsortedfirstvec but returns the last index where v[i] <= x[i].

How it works:

The exponential search starts with a linear scan of the first 5 elements (for cache efficiency), then switches to exponential doubling to quickly narrow the search range before falling back to binary search. This provides O(log k) complexity where k is the distance from lo to the target, instead of O(log n) for full binary search.

The vectorized search functions check if x is sorted and, if so, maintain a monotonically advancing lower bound for subsequent searches. This significantly reduces comparisons when searching for multiple values that are already ordered.

Test plan

• All existing tests pass
• Added tests for searchsortedfirstexp covering basic functionality, edge cases, float vectors, empty/small vectors, repeated elements, and large vectors
• Added tests for searchsortedfirstvec and searchsortedlastvec comparing against element-wise searchsortedfirst/searchsortedlast
• Tests verify correct behavior for sorted and unsorted input vectors
• Tests cover edge cases (values outside range, empty input, single element)

Fixes #8

cc @ChrisRackauckas

🤖 Generated with Claude Code