tts --pronunciation-dictionary-locators for PLS dictionary attachment

[longevityboris]

Context

Flagged by Codex audit. text_to_speech_full and text_to_speech_full_with_timestamps in the OpenAPI spec accept pronunciation_dictionary_locators (array of {pronunciation_dictionary_id, version_id} objects), but tts has no way to send them.

Use case

We already ship the dict family for creating/managing PLS lexicons; without this flag, a voice produced with a dictionary can't actually apply it at TTS time.

What to do

• Add --pronunciation-dictionary <id>[@version] (repeatable) to tts.
• Parse each value into {pronunciation_dictionary_id, version_id?}.
• Forward as pronunciation_dictionary_locators on every TTS endpoint variant (default, stream, with-timestamps, stream+with-timestamps).
• Update TTS_HELP with an example combining dict add-rules → tts --pronunciation-dictionary.

Files

• src/cli.rs (TtsArgs)
• src/commands/tts.rs
• src/help.rs (TTS_HELP + DICT_ADD_RULES_HELP cross-link)
• src/commands/agent_info.rs

Acceptance

• tts "aorta" --pronunciation-dictionary dict_abc@v1 --voice Rachel produces audio that uses the dictionary.
• Integration test asserts the body carries the locator shape correctly.