Fix launch crash on non-string Nightscout treatment timestamps

[zersoft]

Summary

Loop Follow 7.1.0 (60) aborts on every launch when Nightscout returns a treatment whose timestamp field is present but not a string (JSON null or an epoch number). The bolus/SMB/carb/note/stats parsers check != nil and then as! String, so NSNull or NSNumber passes the nil check and traps in swift_dynamicCast on the main queue. The crash repeats on every launch as long as the record is inside the download window.

Crash signature (macOS 27, Designed for iPad build, also reproducible on iOS):

swift::swift_dynamicCastFailure → swift_dynamicCast → Loop Follow (processNS* via updateTreatments, DispatchQueue.main.async)
x24 = "timestam" (ASCII), x28 = value witness table for Date

Changes

• Add NightscoutUtils.treatmentDate(from:): uses timestamp, falls back to created_at, accepts ISO 8601 strings and epoch milliseconds/seconds, returns nil (entry skipped) for anything else.
• Use it in Treatments/Bolus.swift, SMB.swift, Carbs.swift, Notes.swift and the three parsers in Stats/StatsDataFetcher.swift, replacing the forced casts.
• Replace the remaining forced casts of the same class with conditional binding: recommendedTempBasal.timestamp in DeviceStatusLoop/DeviceStatusOpenAPS, predicted.values / prediction.last! in DeviceStatusLoop, and as! TimeInterval / as! String in SAge, CAge, IAge.

Testing

• New Tests/NightscoutTreatmentDateTests.swift (5 cases: string, created_at fallback, null fallback, numeric ms/s, unsupported values → nil).
• Full Tests target: 102 passed, 0 failed (iOS 27 simulator).
• SwiftFormat --lint clean.
• Built and uploaded to TestFlight via the fork's build workflow (CI green). Real-device confirmation against the Nightscout instance that triggered the crash is pending and will be reported here.

🤖 Generated with Claude Code

[zersoft]

Real-device confirmation: the TestFlight build with this fix (fork build of 7.1.0) launches and runs normally on the Mac and Nightscout instance that reproduced the crash on every launch before. No further crashes observed.

[bjorkert]

LGTM, Thanks for your contribution!