feat!: regenerate SDK on fern-typescript-sdk@3.70.1 with TTS-only surface

[luke-speechify]

Regenerated on the current Speechify spec with the latest generator (fern-typescript-sdk@3.70.1). The TypeScript SDK generator image was renamed from fernapi/fern-typescript-node-sdk to fernapi/fern-typescript-sdk at version 0.65.0; this PR adopts the new image name and pins to its current release. Brings the SDK shape in line with industry conventions (apiKey:, flat client.audio.* namespace) and adopts the core.HttpResponsePromise<T> return-type pattern from the 3.x generator line.

Voice-agents methods and types are not exported from this release. The voice-agents API is public over HTTP today, but has never shipped in @speechify/api npm releases — the SDK is deliberately scoped to the stable TTS surface (audio.*, voices.*). Voice-agents methods will be added in a future release once that surface is locked.

What changed

• Auth option renamed. new SpeechifyClient({ token: '...' }) is now new SpeechifyClient({ apiKey: '...' }). The SPEECHIFY_API_KEY environment variable continues to work without change.
• Method namespace flattened. client.tts.audio.speech({...}) → client.audio.speech({...}); client.tts.voices.list() → client.voices.list(); and so on across all six methods. The tts. intermediate namespace has been removed so SDK paths mirror the underlying URL paths (/v1/audio/..., /v1/voices/...).
• Return type changed from Promise<T> to core.HttpResponsePromise<T>. Both are awaitable and resolve to T, but consumers that imported Promise<T> directly from method signatures may need to update type imports. Use .asResponse() or .withResponse() on the returned promise when you need raw fetch Response access.
• Cross-major generator bump (0.49.1 → 3.70.1). Internal code structure has changed substantially. Public method signatures on audio.speech, audio.stream, voices.list / get / create / delete / downloadSample are preserved except for the namespace flatten and the return-type change above.

New in this release

• .fetch() passthrough method on the client for hitting endpoints not yet supported in the typed SDK, while still benefiting from configured auth, retries, logging.
• .withResponse() on every method's returned promise for direct access to the raw fetch Response.
• Better TSDoc on audio.speech and audio.stream documenting the JSON-vs-raw-bytes response distinction between the two endpoints, plus the audio/pcm → audio/L16 Content-Type asymmetry on streaming (24 kHz mono LE PCM).

Release

This PR is the regenerated SDK only. Once merged into master, release-please will open a release PR that bumps the version (to 2.0.0) and updates CHANGELOG.md. Merging the release PR triggers the publish workflow to npm.

BREAKING CHANGE: new SpeechifyClient({ token: ... }) is now new SpeechifyClient({ apiKey: ... }). The tts. namespace has been removed — call sites move from client.tts.audio.speech(...) to client.audio.speech(...) and from client.tts.voices.X to client.voices.X. Method return type changed from Promise<T> to core.HttpResponsePromise<T>; both are awaitable and resolve to T but type-import sites may need updates. Generator-line crossed 0.x → 3.x (plus image rename fern-typescript-node-sdk → fern-typescript-sdk), so internal/private code paths have moved.