M7.9: Text to Speech narrations (panel, tts.* commands, built-in voice)

Write what a narrator says, pick a voice, pitch and pace, and add it to
the timeline. Selecting a narration clip loads it back into the panel;
Save regenerates it in place as one undo step, keeping the clip's start
and length (longer speech kept whole in the file, shorter speech padded
with silence).

- filmcraft-tts (L2): Voice trait, voice catalogue, pause markers
  ([pause 1s]), and an original formant synthesizer (Basic Female/Male;
  robotic placeholder until the neural voices land).
- project: Narration + Project::narrations (schema v13, no-op migration).
- engine: tts.voices/preview/create/edit/inspect; narrations never
  overwrite clips (free targeted track, free track, else a new track).
- ui: Text to Speech panel beside Properties, preview playback,
  click-to-edit, automation ids.

Gates: fmt, clippy -D warnings, layers, assets and wasm (rustup
toolchain) pass. Tests: all pass except three that fail identically on
main (61238bf) on this machine: shortcuts_tests::assign_reassigns_
conflicts_and_undoes, gpu fx_tests::gpu_effects_match_cpu_exactly, and
the frame pool test, which passes when run alone.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/architecture.md

@@ -435,6 +435,10 @@ stereo / mono track ──5.1 panner───┼─► 5.1 submix / 5.1 Mix ─
 
 **Voice-over recording** (`engine::voiceover`, `audio.voiceover.*`). The record point R is the playhead, or the In point when In/Out are set (punch-in; punch-out at Out). Playback starts the pre-roll before it (C = max(0, R − pre-roll)) and the input is captured from C. When the UI's audio clock really starts, `audio.voiceover.sync` restarts the capture there. `audio.voiceover.stop` keeps the audio from R to min(stop, Out), writes a mono 32-bit float WAV `<Name> <n>.wav` (Scratch Disks ▸ Captured, else next to the project, else the data or temporary directory), imports it and overwrites it onto the record track at R as one undo step. The record track is the given one, else the record-armed one, else the first targeted one. Input goes through the `AudioInput` trait: cpal in the desktop app (`apps/filmcraft/src/audio_in.rs`, on its own thread), and `SyntheticInput` headless, which produces exactly the samples the timeline asks for so recordings land sample-accurately. Voice-Over Record Settings (Source, Input channel, Name, Countdown Sound Cues, Pre-/Post-roll) are preferences (`voiceOver`). The track header's microphone records or stops, and a right-click opens the settings dialog. Countdown beeps (1 kHz, 100 ms, each whole second of pre-roll and at R) are mixed into playback, an overlay counts down, and playback stops at Out + post-roll.
 
+### 5.1.2a Text to Speech (narrations)
+
+**Narrations** (`engine::narration`, `tts.*`; voices in `filmcraft-tts`). The Text to Speech panel (Window ▸ Text to Speech) writes a script with pause markers (`[pause 1s]`), a voice, a vocal pitch and a pace; `tts.create` synthesizes it, writes `Narration <n>.wav` (mono 32-bit float, 24 kHz, where voice-over recordings go), imports it and places it at the playhead on the first free targeted audio track, else the first free audio track, else a new one, as one undo step. The settings are kept per generated item in `Project::narrations` (schema v13). Selecting a narration clip loads it into the panel; `tts.edit` writes a new file (beside the previous one) and swaps the clip to it, keeping the clip's start, duration and speed: longer speech is kept whole in the file but plays only to the clip's end; shorter speech is padded with silence to the clip's length. The old item and narration stay, so undo is exact. `tts.preview` synthesizes without changing the project and the UI plays `Session::tts_preview` on the audio output. Built-in voices are an original formant synthesizer (robotic, no download); neural voices come later.
+
 ### 5.1.3 Remix
 
 ```text