Build end-to-end iOS Core ML acceptance

[triasha72]

Summary

• export the tracked current neural checkpoint and preprocessing state into an FP16 Core ML ML Program
• build and test the SwiftUI/Core ML app on an unsigned iOS simulator in macOS CI
• record one cold plus 100 warm Core ML application-level runs
• export provenance-bound physical-iPhone evidence JSON with device, OS, thermal, drift, and latency fields
• validate physical-iPhone evidence and generate JSON/Markdown reports
• retain the simulator app, XCTest result, Xcode version, and checksums in the end-to-end release bundle
• add the iOS lane to the portfolio acceptance matrix and correct public claim boundaries

Local verification

• 19 focused Python tests passed
• Ruff format and lint passed across the repository
• Swift source syntax parsing passed
• GitHub Actions and XcodeGen YAML parsed successfully

Claim boundaries

The CI lane proves Core ML export, app compilation, and XCTest execution on an iOS simulator. It does not claim physical-iPhone latency, Apple Neural Engine placement, or power. The app's physical-device JSON is accepted only when the repository validator confirms a non-simulator iPhone, at least 100 warm runs, matching model/preprocessing hashes, and output drift within tolerance. ANE and power remain not_measured unless separate named Instruments/measurement evidence is retained.