Add Pre/Off/Post EQ placement (closes #478)

[silvansky]

Closes #478.

Adds a Pre / Off / Post control for the tone stack (EQ), so the EQ can run before the NAM model, after it, or be bypassed. The blocker noted in #478 (#459 — legacy unserializer) has since been resolved, and this PR routes the new parameter through that existing migration framework so old projects keep loading correctly.

What changed

• Param: merged the old kEQActive bool into a single 3-state enum kEQMode {"Pre","Off","Post"} (default Post, i.e. the current behavior). Enum EEQMode { kEQModePre, kEQModeOff, kEQModePost }.
• DSP (ProcessBlock): in Pre mode the tone stack processes the signal feeding the model (gate detection stays on the raw input); in Post mode it runs after the model as before; Off bypasses it. The tone stack is invoked exactly once (Pre XOR Post) and writes to its own buffer (no input/output aliasing).
• UI: replaced the EQ slide switch with an IVTabSwitchControl (Pre/Off/Post) plus a label, in the same spot. EQ knobs grey out in Off.
• Backward compatibility (Unserialization.cpp, version 0.8.0 → 0.8.1): new _GetConfigFrom_0_8_1 reader (param renamed ToneStack → EQ), and a migration in _UpdateConfigFrom_0_7_14 mapping the legacy bool (on → Post, off → Off). Because the older readers chain through _UpdateConfigFrom_0_7_14, presets from all prior versions — including current 0.8.0 — migrate correctly; new presets round-trip via the 0.8.1 reader.

Legacy presets can only map to Post (pre-model EQ didn't exist before) — intended.

On the version slot

Rebased onto current main (0.8.0), and the new reader is gated at _Version(0, 8, 1) rather than a patch of the 0.8.x line that already exists.

This matters, because the serialized chunk stores unnamed ordered doubles: if the EQ reader were gated at any version ≤ 0.8.0, every preset written by the released 0.8.0 would be read with slot 8 (ToneStack, a bool) reinterpreted as the EQ enum — 1.0 (tone stack on) → kEQModeOff, 0.0 (off) → kEQModePre. Inverted, and the ToneStack → EQ migration would never run for that range. Gating at 0.8.1 keeps 0.7.14 … 0.8.0 on the ToneStack reader where the migration applies.

Happy to renumber if you'd prefer a different version slot — 0.8.1 is just the next patch; if you have your own 0.8.1 planned with different serialization, say the word and I'll move it.

Notes

• The 7 *-Info.plist changes and the installer/NeuralAmpModeler.iss bump are auto-generated from the config.h version bump, via this repo's own scripts/update_version-mac.py, update_version-ios.py and update_installer-win.py.
• I did not run format.bash: my local clang-format (v22) reformats unrelated pre-existing code (collapsing short brace-bodies per the repo's .clang-format), which would add churn. It also wants to re-wrap two of the lines this PR adds — happy to apply whichever clang-format version you consider canonical if you'd like the added lines normalized.

Testing (per CONTRIBUTING)

• The standalone plugin builds.
• The plugin runs.
  • The plugin makes sound.
  • You can load a new-style (file) model.
  • You can load an old-style (directory) model.
  • You can load a supported IR.
  • The EQ section works (Pre/Off/Post switch + knobs; Pre vs Post audibly changes tone, Off bypasses).
• The VST3 plugin builds.
  • The plugin passes validation. Ran pluginval 1.0.4 @ strictness 8 (the tool this repo's CI uses) against the VST3 → SUCCESS, 0 failures. Covers Plugin state + Plugin state restoration (exercises the new serialization/migration), Audio processing and Non-releasing audio processing across 44.1/48/96 kHz × block sizes 64–1024, Automation, Parameters, Parameter thread safety, and Fuzz parameters.
• The AU plugin builds.

[silvansky]

@sdatkinson any thoughts?