Speed up PCS waveform evaluation

[nicamo]

Summary

Reduce repeated interpolation overhead in the PCS control-loop hot path without changing waveform semantics or public calling syntax.

Changes

• add a lightweight ConstantInterpolant for waveforms that are constant in time
• use it for constant step and spline waveforms while preserving scalar and array output shapes and derivative compatibility
• construct spline derivatives whenever controller interpolants are rebuilt
• reuse cached derivatives in shape, systems, and virtual-circuit controller evaluations
• add focused regression tests for constant waveform output shapes and derivative reuse

Motivation

PCS controllers may evaluate the same interpolants many times between equilibrium solves. Previously, derivative objects were reconstructed on every evaluation, and constant schedules still passed through general SciPy interpolators.

Performance

Focused microbenchmarks on this branch gave:

• cached spline derivative evaluation: about 5.9x faster
• constant step waveform evaluation: about 3.4x faster
• constant spline evaluation: effectively unchanged

The expected end-to-end scenario gain is modest because Grad-Shafranov solves dominate runtime. The benefit is localized to PCS evaluation and grows when many PCS substeps are performed per equilibrium step.

Compatibility

• no public API changes
• non-constant waveforms retain the existing SciPy interpolation path
• constant interpolants preserve scalar and array output shapes
• cached derivatives are rebuilt whenever interpolants are updated

Validation

python -m pytest -q freegsnke/tests/test_control_loop_interpolation.py

Result: 3 passed.

[kpentland]

Very nice - squeezing a little bit more efficiency!