Save computed normalization stats (`audio_mean`, `audio_std`, `spec_mean`, `spec_std`) to JSON

[riley-1995]

Context

Teammates need access to normalization statistics produced by the current preprocessing workflow.
These stats are currently computed in code but not persisted in a standard artifact.

Goal

Persist normalization stats to a JSON file as part of preprocessing so downstream users/scripts can reliably consume them.

Requirements

• Save the following keys in JSON:
  • audio_mean
  • audio_std
  • spec_mean
  • spec_std
• Use current preprocessing pipeline outputs (no Audio-Processing integration dependency for this issue).
• Ensure JSON is written deterministically to a documented location.
• Update README with:
  • where file is written
  • how it is generated
  • example JSON snippet
• Keep this change minimal and mergeable quickly.

Suggested Output Path

• data/normalization_stats.json
  (or another repo-consistent path; document final choice in README)

Acceptance Criteria

• Running preprocessing produces a JSON file with all 4 required keys.
• Values are numeric and non-null when data exists.
• README documents generation + location.
• PR includes example output in description/log.

Out of Scope

• Audio-Processing submodule integration
• Refactoring preprocessing architecture
• Broader training pipeline redesign