chore: refactor setup steps

[goodroot]

It was getting a little messy. +, real time is no longer experimental.

[github-actions]

Continue PR Review

Automated review via Continue CLI run through opub.

Summary

This refactor is mostly cosmetic, but it introduces one likely functional regression in realtime setup and one compatibility risk around custom realtime endpoints. Test coverage should also be extended for the new realtime selection flow.

Findings

• Realtime setup no longer allows selecting converse mode, changing behavior for existing/new users. In lib/src/cli_commands.py, the previous realtime_mode prompt for OpenAI/Google/custom backends was removed and replaced with remote_config['realtime_mode'] = 'transcribe'. The docs still advertise converse, but setup now silently forces transcribe and only mentions editing config later. This is a user-visible regression in setup flow and can break expectations for anyone reinstalling/reconfiguring realtime backends.
• Custom realtime backends may be forced into the wrong URL shape. _prompt_realtime_provider_model_selection() lets users enter a full custom websocket_url, but WhisperManager._get_websocket_url() builds provider URLs differently depending on mode (?intent=transcription vs ?model=...). If custom endpoints are passed through _generate_remote_config and later reused by _get_websocket_url, appending provider-specific query params may corrupt already-complete custom URLs. This needs an explicit test/contract for custom realtime URLs to ensure the entered URL is used verbatim or merged safely.
• Provider registry removal of gpt-realtime-mini-2025-12-15 can break existing saved configs. lib/src/provider_registry.py removes that model from OpenAI’s registry. If users already have websocket_model: gpt-realtime-mini-2025-12-15 in config, any validation/reconfiguration path that assumes the model must exist in the registry may now fail or downgrade unexpectedly. At minimum, add a migration/compatibility path or preserve deprecated models as hidden entries until config migration is handled.

Suggested follow-ups

• Add setup tests covering realtime backend configuration for OpenAI/Google/custom, including preserving/selecting realtime_mode.
• Add tests for custom realtime endpoints to verify the configured websocket_url is not modified incorrectly at connection time.
• Add a compatibility test for existing configs using deprecated realtime model IDs, or keep removed models as hidden deprecated entries until migrated.