feat(ios): Implement Native Classifier with Accelerate & BNNS

[StoyanD]

Rationale

To support high-performance on-device content classification in the Alete iOS app without the overhead of a JS runtime.

Changes

• New AleteClassifier Swift Package.
• Native BertTokenizer implementation.
• BNNS-accelerated MLP head.
• vDSP-accelerated embedding pooling and normalization.
• Int4 Dequantization support.
• New Conductor track for native implementation tracking.

Testing Performed

• Full unit test coverage for tokenization and model loading.
• Verified parity against JS implementation with 100+ diverse samples using parity_data.json.
• Performance benchmarks conducted on iPhone 15 Pro showing significant latency improvements.

[StoyanD]

@copilot resolve the merge conflicts in this pull request

[Copilot]

@copilot resolve the merge conflicts in this pull request

Resolved. The only conflict was in PROGRESS_REPORT.md where both branches had a "Section 9" entry. I kept the npm Package Split as section 9 (since it was merged to main first) and renumbered the iOS Native Classifier as section 10. Merge commit: 0dea00e.