Fix complementary-filter quaternion interpolation threshold

[Daniiiil1]

What changed

• use the published 0.9 scalar-component threshold when choosing between SLERP and normalized LERP
• keep SLERP for corrections at or below the threshold and LERP only for small corrections above it
• add numerical tests for both interpolation branches
• wire the new test into the ament test configuration

Why

The complementary-filter paper defines epsilon as 0.9: when the delta quaternion scalar component is greater than epsilon, the correction is small enough for normalized LERP; otherwise the filter uses SLERP. The implementation currently compares against zero, which sends most non-negative, non-small corrections through LERP instead.

Fixes #217.

Reference: Valenti, Dryanovski, and Xiao, Keeping a Good Attitude: A Quaternion-Based Orientation Filter for IMUs and MARGs, equations 48–52.

Validation

• git diff --check
• Apple Clang 21 compile of complementary_filter.cpp with C++17 and -Wall -Wextra -Wpedantic
• standalone numerical probe verifying the exact SLERP result below the threshold and normalized-LERP result above it
• clang-format --dry-run --Werror on the changed C++ files

A full ament test run was not available locally because this machine does not have a ROS 2 installation; the added gtest is intended to run in the repository CI.