Integrate `Audio-Processing` submodule into `ELP-CNNvsRNN-v2` preprocessing/training workflow

[riley-1995]

Context

Audio-Processing submodule has been added but not yet integrated into the active workflow.
Current pipeline still relies on legacy/local implementations.

Goal

Adopt submodule functionality in a controlled migration to reduce duplication and standardize DSP/pipeline behavior.

Proposed Scope

• Map existing preprocessing steps to Audio-Processing APIs.
• Define adapter layer and path/config compatibility.
• Replace selected local preprocessing components incrementally.
• Preserve current outputs/shape/contracts needed by training.
• Update docs and commands for the new workflow.

Milestones / Checklist

• Inventory overlap between src/elp_rumble/* preprocessing and Audio-Processing/src/audio_processing/*
• Define migration plan (what moves first, what stays temporarily)
• Add compatibility/adaptor module(s)
• Migrate one stage at a time (I/O, DSP, pipeline orchestration)
• Verify parity of generated artifacts (shape/labels/distribution checks)
• Update README and operational scripts
• Add tests/smoke checks for migrated path
• Remove deprecated duplicate code where safe

Risks

• Behavioral drift in generated TFRecords/spectrograms
• Path/config mismatch between current project and submodule conventions
• Longer review cycle if done in one large PR

Delivery Strategy

Use multiple smaller PRs linked to this issue rather than one monolithic PR.