feat(learn,todo,meditate): dictate through the phone's voice input app (ADR 0038, D18)

On Android the microphone now opens the app handling
RECOGNIZE_SPEECH (FUTO Voice Input, Google's), the default the person
chose in Android, and puts the text it returns where the cursor stood,
as Dictation does; cancelled, nothing. Only where none is installed, or
it can't be opened, does speech_to_text listen in line as before.

halkyon_app: VoiceInputApp (one listen, an answer once closed) on the
channel app.halkyon/voice_input, which DeviceSpeechInput asks first at
every listen, answering SpeechStart.finished so that Dictation doesn't
listen again; speech_to_text isn't touched on that route, so our app
asks for no microphone. SpeechInput.listen and Dictation.start take the
prompt the voice input app shows (the field's hint).

Each app: VoiceInput.kt (queryIntentActivities, startActivityForResult,
the first of EXTRA_RESULTS) attached in MainActivity, which hands it
activity results, and RECOGNIZE_SPEECH among the manifest's queries.
The privacy pages of Learn and Meditate say so.

Our own Whisper and Meditate's dream recordings are dropped (D18).

Whats-New: The microphone now dictates through your phone's voice input app, such as FUTO Voice Input.
Whats-New-De: Das Mikrofon diktiert jetzt über die Spracheingabe-App deines Telefons, etwa FUTO Voice Input.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>

docs/adr/0038-halkyon-meditate.md

@@ -204,6 +204,39 @@ names and pictures are its own). Everything below is Halkyon's own.
   - The platforms declare it as Learn's do: `RECORD_AUDIO` and the recognition service's
     query on Android, the microphone's and speech recognition's usage texts on iOS and
     macOS, and macOS's `audio-input` entitlement.
+- **D18 Dictation through the phone's voice input app** (7 Nov 2026, the developer's
+  decision, amending D15 for Learn's, Todo's and Meditate's dictation alike): on Android
+  the microphone opens the app handling `RecognizerIntent.ACTION_RECOGNIZE_SPEECH` (FUTO
+  Voice Input, or Google's), whichever the person made the default in Android itself, and
+  puts the text it returns where the cursor stood, as D15's dictation does (halkyon_app's
+  `DictationInsertion`); cancelled, nothing. Only where none is installed, or it can't be
+  opened, does D15's in-line recogniser (`speech_to_text`) listen; where neither exists the
+  microphone stays hidden. Text only: no sound is kept, and our app asks for no
+  microphone on this route, as the voice input app records.
+  - **Why**: FUTO Voice Input recognises on the device with Whisper, but offers no
+    `SpeechRecognizer` (`RecognitionService`) API, registering only a placeholder
+    service, so `speech_to_text` can't reach it; it does handle the intent, in a floating
+    window, and returns text. Our own Whisper (whisper.cpp in a package of ours, the
+    `whisper-spike` branch) is not pursued: a model to download and a native build per
+    CPU, for what an installed app already does. FUTO's code is under the FUTO Source
+    First License, which doesn't go with our AGPL, so it is used as a separate app only,
+    never linked or copied. And the dream's recordings (audio kept beside the text,
+    synced end to end encrypted in a table of their own) are dropped: the transcript is
+    enough, and the intent returns text, not audio.
+  - **How**: `VoiceInputApp` in halkyon_app (one listen, an answer once closed: its own
+    shape, as `SpeechInput` streams words and `Dictation` listens again after each pause),
+    on each app's `VoiceInput.kt` over the channel `app.halkyon/voice_input` (`installed`
+    by `queryIntentActivities`, `listen` by `startActivityForResult`, the first of
+    `EXTRA_RESULTS`), with `EXTRA_LANGUAGE_MODEL` free form, `EXTRA_LANGUAGE` the app's
+    language and `EXTRA_PROMPT` the field's hint (Meditate: *What did you dream?*); each
+    manifest declares the intent among its `<queries>` (Android 11 on). `DeviceSpeechInput`
+    asks it first at every tap, answering `SpeechStart.finished`, after which `Dictation`
+    doesn't listen again. While it is open the button shows Stop and nothing else listens;
+    Stop drops what it returns later.
+  - **Consequences**: on a phone with Google's app, which handles the intent too, its
+    dialog listens now rather than the in-line recogniser, one utterance a tap, and D15's
+    recognition on the device (`preferOnDevice`) applies only where no app handles the
+    intent; which recognises on the device is then that app's choice.
 
 ## The operator's library (3 Nov 2026, the developer's request)