fix(tremolo): perceptual shape curve so the whole knob morphs

[OpenSauce]

Problem

Follow-up to #258. The Shape knob "doesn't really change much" — and the numbers confirm it. Shape morphs the LFO from sine → hard square via tanh waveshaping, but that morph saturates with drive, and drive was mapped linearly (0 → 12). Result: ~80% of the morph happens in the bottom third of the knob, and the entire upper half (shape 0.5 → 1.0) only crawls from 80% → 90% square. The top of the knob is effectively inert.

(Shape is also inherently subtle at low Depth — it's a volume effect; the sine-vs-square character only gets dramatic near full depth. That part is expected.)

Fix

The morph saturates with drive, so equal knob steps should map to equal ratios of drive — a geometric sweep, not a linear one. Drive now ramps geometrically from SHAPE_MIN_DRIVE (0.5) to MAX_DRIVE (raised 12 → 16 for a crisper top-end square), with a true sine preserved at exactly shape = 0. Still computed only on shape change, off the RT path.

Measured squareness across the knob (shape 0..1 in eighths):

before (linear):  0  29  59  73  80  84  86  88  90    step σ 11.1
after  (geom.):   0  10  21  38  56  72  82  88  92    step σ  4.7

Now the morph is spread evenly — shape 0.5 sits at 56% (was 80%), so the upper half of the knob actually does something.

Tests

Adds upper_shape_knob_still_morphs: the shape 0.5 vs 1.0 LFO envelopes must differ by MAE > 0.03 (linear map ≈ 0.019 → would fail; geometric ≈ 0.066 → passes). All existing tremolo tests still pass (true sine at shape 0, full-chop silence/unity at shape 1, gain in [0,1], periodicity, cache correctness).

Verification

• cargo clippy --workspace --all-targets -D warnings -D pedantic -D nursery — clean
• make test — 164 core + 30 no-alloc pass (11 tremolo)

[Copilot]

Pull request overview

This pull request adjusts the tremolo stage’s Shape control mapping so the sine→square morph is distributed more evenly across the knob’s full range, addressing the previously “inert” upper half of the control.

Changes:

• Replace the shape→drive mapping from a linear sweep to a geometric sweep, with a special-case at shape == 0 to preserve a (near) pure sine.
• Increase MAX_DRIVE (12 → 16) to achieve a crisper near-square at full shape.
• Add a regression test ensuring the upper half of the Shape knob still produces meaningfully different LFO envelopes.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.