refactor: use urllib.parse for WebSocket URL construction

[kraenhansen]

Summary

• Extract a shared build_ws_url(base_url, path_segments, params) helper in url_utils.py that handles:
  • Scheme conversion (http→ws, https→wss)
  • Percent-encoding each path segment (preventing injection via dynamic values like voice IDs)
  • Query string encoding via urllib.parse.urlencode
• Replace manual f-string URL construction in all three WebSocket URL builders:
  • ScribeRealtime._build_websocket_url in scribe.py
  • RealtimeTextToSpeechClient.convert_realtime in realtime_tts.py
  • BaseConversation._get_wss_url in conversation.py
• Fix _get_signed_url to use urlparse/urlunparse instead of string concatenation for appending query params

Closes #779

Test plan

• All 16 tests in test_url_utils.py pass (scheme conversion, path encoding, special chars, query params)
• All 13 tests in test_stt_realtime.py pass (including URL construction and scheme conversion)
• All 19 tests in test_convai.py + test_async_convai.py pass (including URL edge cases)

🤖 Generated with Claude Code

---

Note

Medium Risk
Touches WebSocket URL generation for conversation, realtime STT, and realtime TTS; subtle changes in path/query encoding or base-path joining could break connectivity in edge cases.

Overview
Introduces a shared build_ws_url helper to construct WebSocket URLs by converting schemes (http(s)→ws(s)), percent-encoding path segments, and encoding query parameters via urllib.parse.urlencode.

Refactors the conversation, realtime STT (ScribeRealtime), and realtime TTS URL builders to use this helper instead of manual string/urljoin construction, and updates conversation signed-URL augmentation to append source/version via urlparse/urlunparse.

Adds unit tests covering scheme conversion, path joining/encoding, query encoding (including repeated keys), and port/base-path handling.

^{Reviewed by Cursor Bugbot for commit 3b97a1f. Bugbot is set up for automated code reviews on this repo. Configure here.}