Fix spring-damper damping regime scaling

[Vedangalle]

Summary

• Scale the characteristic-discriminant tolerance by the larger discriminant term.
• Add analytical regression coverage for overdamped, critically damped, and underdamped responses.
• Verify time-unit invariance in all three damping regimes.

Fixes #3521.

Problem

mju_springDamper previously compared Kv^2 - 4*Kp directly with the fixed absolute value mjMINVAL. A consistent time rescaling multiplies the discriminant by the square of the time-scale factor, so this comparison could change the selected damping regime even though the physical trajectory was unchanged.

For one equivalent overdamped pair, the unscaled call returned 0.7946142382307649, while the time-scaled call returned 0.0404276819945128. Both should return 0.7946142382307650.

The updated tolerance is relative to max(Kv^2, 4*abs(Kp)). It therefore scales with the discriminant and retains the near-critical branch without changing the public API.

Validation

• New spring-damper regression tests: 2/2 passed.
• Full engine_util_misc_test binary: 66/66 passed.
• Full CTest suite: 1376/1376 passed.
• Independent property sweep: 1080 equivalent parameterizations passed across all damping regimes and time scales from 1e-2 through 1e-12; maximum absolute difference was 5.33e-15.

Local environment: macOS, AppleClang 21, CMake 4.4.2, Release build with MUJOCO_HARDEN=ON.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.