[MR V2] Migrate to Model Runner V2

Squashed migration rebased onto current main. Collapses the 46-commit
dev/migrate-MR-v2 history (10 merge commits + superseded pre-rebase work)
into a single commit, so the diff reflects only the genuine MR-V2 feature
(worker_v2/ runners + model_states, omni schedulers, stage engine pipeline,
pooling_output serialization bridge). Content already merged to main
separately (step_audio2, tts_adapters, npu/_310p, etc.) drops out.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/helpers/assertions.py

@@ -541,6 +541,15 @@ def assert_audio_speech_response(response: Any, request_config: dict[str, Any],
 
     assert response.success, "The request failed."
 
+    # Optional floor on decoded audio size (models with very short clips may use a lower value).
+    min_audio = request_config.get("min_audio_bytes")
+    if min_audio is not None:
+        n = int(min_audio)
+        if n > 0:
+            ab = response.audio_bytes
+            assert ab is not None, "Expected audio bytes when min_audio_bytes is set"
+            assert len(ab) > n, f"Audio payload too small: {len(ab)} bytes, expected more than {n} (min_audio_bytes)"
+
     # Optional floor on decoded audio size (models with very short clips may use a lower value).
     min_audio = request_config.get("min_audio_bytes")
     if min_audio is not None: