Add support for embedding NAM models and IR files in session data

[damiangr]

Description

This PR adds the ability to embed NAM model files (.nam) and impulse response files (.wav) directly into DAW session/project files, making sessions fully portable and self-contained.

Features

• Automatic Embedding: When saving a session, NAM models and IR files are encoded and embedded in the plugin state
• Automatic Extraction: When loading a session, embedded files are extracted to temporary files and loaded automatically
• Backward Compatible: Sessions without embedded data continue to work with file paths as before
• Cross-Platform: Designed to work on Windows, macOS, and Linux

Benefits

• Share sessions without worrying about missing model/IR files
• Old projects remain playable even if original files are moved/deleted
• Team members can open sessions without manual file management

Implementation

• Modified SerializeState() to embed file contents in session data
• Modified UnserializeState() to detect and extract embedded files automatically
• Added _StageModelFromData() and _StageIRFromData() helper methods for loading from memory
• Maintain backward compatibility with file path references
• Update AudioDSPTools submodule to include SIGFPE fix (depends on sdatkinson/AudioDSPTools#24)

PR Checklist

• [x] Did you format your code using format.bash?
• [x] Does the VST3 plugin pass all of the unit tests in the VST3PluginTestHost?
  • [x] Windows - Validated with pluginval (strictness level 5) - SUCCESS
  • [ ] macOS - Unable to test: VMware Sonoma 14.7.2 + Xcode 16.2 environment has NanoVG/Metal incompatibility issues. Recommend testing via GitHub Actions CI which uses Xcode 15.x.
• [x] Does your PR add, remove, or rename any plugin parameters? No
• [x] Does your PR add or remove any graphical assets? No

Testing Notes

Windows (Validated ✅)

• Build: Visual Studio 2022, x64 Release
• Validation: pluginval v1.x, strictness level 5
• Results: All tests passed (audio processing, automation, state save/restore, editor, bus configuration)

macOS (Limited Testing ⚠️)

Note: Local macOS testing was not possible due to environmental constraints:

• Test environment: VMware Sonoma 14.7.2 + Xcode 16.2
• Issue: NanoVG/Metal incompatibility causing image loading crashes (not related to this PR)
• Official v0.7.13 release also crashes in this environment
• Recommendation: Please test via GitHub Actions CI, which uses Xcode 15.x and should work correctly

The embedding code itself is platform-independent and uses standard C++ file I/O. The implementation should work on macOS, but comprehensive validation via CI is recommended.

Dependencies

• Depends on sdatkinson/AudioDSPTools#24 (SIGFPE crash fix)

Files Changed

• AudioDSPTools: Updated submodule to include SIGFPE fix from PR #24
• NeuralAmpModeler.cpp: Embedding/extraction logic and helper methods (333 lines added)
• NeuralAmpModeler.h: Added mNAMData and mIRData member variables
• Unserialization.cpp: Added embedded data extraction logic (75 lines added)