Mobile voice input: better transcription options for non-English speakers

[roprgm]

iPhone voice input works poorly for anyone dictating in a language other than English. These are two proposed improvements, from the smallest to one for users who need more accuracy:

1. Choose the transcription language for on-device transcription. It's a configuration option for an existing capability, which the contributing guide allows without prior approval, so it's open as a PR: feat(mobile): choose the language for iPhone voice input #17692
2. Allow users to set a custom transcription service, for those who need more accuracy than on-device transcription gives, even with the right language. It's a new capability, so it waits for approval here before going upstream. It's ready in my fork: feat(mobile): transcribe voice input with a cloud service set up in Settings roprgm/t3code#1

1. Choose the transcription language

iPhone voice input always transcribes in the app's language. The app ships only in English, so iOS reports an English locale even on phones set to another language, and speech in any other language comes out as unrelated English words (#12743).

A Language setting in Settings → Voice input. Automatic stays the default and keeps today's behavior; picking a language makes on-device transcription use it.

Settings	Voice input

2. Allow users to set a custom transcription service

The same screen could also let users bring their own transcription model: an endpoint, model, and API key of their choice. The key stays in the iPhone's Keychain and is only sent to that service, and nothing changes on the server. Users who find on-device transcription falls short can pick any model they want, including cloud models that handle technical terms and mixed languages much better. It also brings voice input to iPhones without on-device transcription.

Settings	Voice input

Allowing users to set a custom transcription service is a new capability, so I won't open it upstream unless this direction is approved.