[endrebak]

Complement can emit an empty or reversed terminal gap: [90,100) on a chromosome of length 100 returns [100,100), or [105,100) with slack 5. Positive slack also moves the boundaries of surviving gaps: [10,20) and [30,40) produce [25,30) and [45,100) instead of [20,30) and [40,100).

Sweep the original coordinates for nonnegative slack and use slack only as a gap-length threshold. Both chromosome-finalization paths omit terminal gaps that are empty, reversed, or no larger than slack. Saturating comparisons also avoid the previous end + slack overflow near the coordinate maximum. Optional leading gaps and negative-slack end contraction retain their existing behavior.

Bump the crate, lockfile, and changelog to version 0.1.15.

This is a separate core follow-up to ruranges_py#44, whose implicit chromosome-ID correction can be merged independently. No public signatures or shared event-builder behavior change.

Validation:

• cargo test --offline: 61 existing unit tests, 14 new integration tests, and 3 doctests pass (78 total).
• Before the fix, 10 new targeted regressions fail; the 2 negative-slack controls pass. The added tests cover i32/i64, both chromosome-finalization paths, exact slack thresholds, nesting, attribution, and coordinate limits. The independent interval-union oracle checks another 6,000 deterministic cases.
• A clean release Python wheel built against this core and the exact #44 dispatcher passes 30 Python checks: all 10 #44 regressions and 20 terminal-gap/slack cases covering implicit/explicit groups and i32/i64.
• rustfmt --check src/complement_single.rs tests/complement_gaps.rs and git diff --check pass. The full formatting check reports an existing formatting difference in src/ranks.rs.
• cargo clippy --offline --all-targets succeeds with existing warnings. The strict -- -D warnings check fails on pre-existing diagnostics in unrelated files; neither changed Rust file has diagnostics.