feat: v0.2.0 deep-dive 2-band -> 3-band rebuild + M2 preset system

[yves-vogl]

Summary

Research-driven v0.2.0 rebuild (docs/design-brief.md, docs/research-notes.md) restructuring Crypta's signal path from a 2-band (low/high) split to a genuine 3-band (low/mid/high) split matching the reference plugin class's own defining architecture, plus the suite-wide M2 preset system.

Topology (headline change)

• Two cascaded LR4 crossovers replace the single split: Split Low (60–400 Hz, default 120 Hz, was Crossover Frequency) peels off the Low band; Split High (300–2000 Hz, default 600 Hz, NEW) splits the remainder into Mid and High. The two are clamped ≥1/3 octave apart (src/dsp/SplitGap.h) to prevent a degenerate near-zero-width Mid band.
• New Mid band (src/dsp/MidBand.h/.cpp): staged/cascaded drive-only saturation (0–100%, default 30%) + Level, no filter/tone/blend — matching the reference class's own Mid band exactly.
• High band's pre-drive highpass (Tight, 20–500 Hz, default 100 Hz) promoted from a Razor-only fixed 200 Hz constant to a first-class, voicing-independent control.
• Low-band compressor ballistics re-sourced to the reference class's fixed "glue" values: ratio 4:1→2:1, attack 10→3 ms, release 120→6 ms (range floor lowered 10→5 ms). Retires the "New York style" framing.
• IR loader relocated to process only the Mid+High post-sum signal, never the Low band — structurally enforced and directly tested (tests/LowBandIsolationTests.cpp).
• EQ default corner frequencies re-anchored (80/500/2800/5000 Hz, sourced bass tone-stack anchors).
• src/dsp/PhaseAlignFilter.h (unplanned engineering necessity, discovered during this rebuild): a naive cascade of two independent LR4 crossovers does not flat-sum on its own — up to −10 dB deviation was measured at close split ratios. Fixed with an allpass compensation filter on the Low band; the fix is proven algebraically (not just empirically) in docs/architecture.md's "Cascaded 3-band flat-sum and phase alignment" section.
• Lossy v0.1.x state migration: the old crossoverFreq value maps to the new splitHighHz, clamped into 300–2000 Hz. v0.1.x's shipped default (250 Hz) lands exactly at the 300 Hz floor — dedicated regression test in tests/StateMigrationTests.cpp.

M2 preset system (ported from basilica-audio/nave's pilot)

• src/presets/PresetManager.{h,cpp}, PresetBar.{h,cpp}, Localisation.{h,cpp} — copied verbatim per the replication recipe.
• Nine factory presets (presets/factory/*.json, documented in docs/presets.md): Default, Glue & Grind, Sub Lock, Throat, Fuzz Wall, Cut Through, Definition Only, Clean Low/Loud Top, Cab-Colored Grind.
• German localisation of the preset UI frame (resources/i18n/de.txt); core/DSP terminology is never translated.
• CMakeLists.txt: juce_add_binary_data(CryptaBinaryData ...), links juce_gui_basics + CryptaBinaryData into SharedCode.
• PluginEditor now docks a PresetBar above the existing generic editor.

Version

Bumped to 0.2.0.

Test plan

• ctest --test-dir build --output-on-failure — 96/96 Catch2 tests green locally (Debug, native arm64), up from 53 pre-rebuild. All v0.1.1 regression coverage (reset(), midFilter reset, tail length, oversized-block chunking) stays green.
• New coverage: SplitGapTests, MidBandTests, ThreeBandFlatSumTests (3-band flat-sum incl. the minimum-gap clamp boundary), StateMigrationTests (incl. the untouched-v1-default → exactly-300Hz test), LowBandIsolationTests (IR-routing guarantee), TopologyRobustnessTests (NaN/Inf sweep of every new/changed control), PresetManagerTests (17 cases, ported from nave).
• cmake --build build --target Tests Crypta_Standalone — both build clean (Debug, Ninja, native arm64).
• CI (macOS + Windows, pluginval strictness 10, auval) — pending, watching after this PR opens.

🤖 Generated with Claude Code