feat: Improve SuperTonic voice resolution and add tests for voice formula handling

tests/test_voice_formula_resolution.py

@@ -0,0 +1,18 @@
+from __future__ import annotations
+
+from abogen.web.conversion_runner import _resolve_voice, _supertonic_voice_from_spec
+from abogen.tts_supertonic import DEFAULT_SUPERTONIC_VOICES
+
+
+def test_resolve_voice_formula_without_pipeline_does_not_crash() -> None:
+    # This can happen when a previously-saved Kokoro mix formula is present
+    # but the active provider is SuperTonic (no Kokoro pipeline object).
+    formula = "af_heart*0.5+af_sky*0.5"
+    resolved = _resolve_voice(None, formula, use_gpu=False)
+    assert resolved == formula
+
+
+def test_supertonic_voice_from_formula_falls_back_to_valid_voice() -> None:
+    # When a stale Kokoro mix formula is present, SuperTonic should not receive it.
+    chosen = _supertonic_voice_from_spec("af_heart*0.5+af_sky*0.5", "af_heart*1.0")
+    assert chosen in DEFAULT_SUPERTONIC_VOICES