feat!: regenerate Python SDK (TTS audio + voices); hoist errors/types to package root

[luke-speechify]

Regenerates the Python SDK from the current API spec. This PR is only the TTS regeneration — the realtime connect layer and the agent namespace that appeared on earlier revisions of this branch are intentionally out of scope here and are tracked separately.

Generated surface

client.tts (and the AsyncSpeechify equivalent) exposes two sub-clients:

client.tts.audio

• speech(*, input, voice_id, audio_format=, language=, model=, options=, request_options=) -> GetSpeechResponse
• stream(*, accept, input, voice_id, language=, model=, options=, request_options=) -> Iterator[bytes]

client.tts.voices

• list(*, request_options=) -> List[GetVoice]
• create(*, name, gender, sample, consent, locale=, avatar=, request_options=) -> CreatedVoice
• delete(id, *, request_options=) -> None
• download_sample(id, *, request_options=) -> Iterator[bytes]

Breaking changes (feat!)

Relative to master:

• Shared errors and types moved to the package root. speechify.tts.errors -> speechify.errors, and speechify.tts.types -> speechify.types. Consumer imports must be updated.
• Agent / telephony modules removed from the generated client. client.tts previously also exposed auth, agents, tools, conversations, knowledge_bases, memories, phone_numbers, sip_trunks, outbound_calls, and workspaces. The regenerated client exposes only audio and voices.

Verification

• CI: compile and test both pass.

Follow-ups

• A possible re-namespace of tts -> audio (client.audio.*) is under discussion and is not part of this PR.

[linear-code]

DRG-31

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.