fix: forward billing-v2 protocol on textchat KB retrieval and node summaries (#503)

Two call sites dropped the MPS billing-v2 protocol, so orgs on model
config v2 got 400 "Service Key uses billing v2" from MPS:

- text_chat_runner built PipecatEngine without embeddings_provider
  (extracted but never passed), so knowledge-base retrieval fell through
  to the plain OpenAI-compatible client instead of DograhEmbeddingService
  and sent no correlation_id/mps_billing_version metadata. Also pass
  endpoint/api_version for Azure BYOK parity with run_pipeline.
- node_summary built its QA LLM without the run's correlation id, unlike
  its sibling call in qa/analysis.py.

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

api/services/workflow/qa/node_summary.py

@@ -7,6 +7,7 @@ from pipecat.processors.aggregators.llm_context import LLMContext
 
 from api.db import db_client
 from api.db.models import WorkflowRunModel
+from api.services.managed_model_services import get_mps_correlation_id
 from api.services.pipecat.service_factory import create_llm_service_from_provider
 from api.services.workflow.dto import NodeType, QANodeData
 from api.services.workflow.qa.llm_config import resolve_llm_config
@@ -75,7 +76,15 @@ async def ensure_node_summaries(
         logger.warning("No API key for node summary generation, skipping")
         return existing_summaries
 
-    llm = create_llm_service_from_provider(provider, model, api_key, **service_kwargs)
+    # Reuse the run's MPS correlation id (minted at run start, persisted on
+    # initial_context) so managed-model-services calls carry billing-v2
+    # markers — orgs on billing v2 reject managed calls that lack them.
+    mps_correlation_id = get_mps_correlation_id(
+        getattr(workflow_run, "initial_context", None)
+    )
+    llm = create_llm_service_from_provider(
+        provider, model, api_key, correlation_id=mps_correlation_id, **service_kwargs
+    )
 
     updated_summaries = dict(existing_summaries)
 

api/services/workflow/text_chat_runner.py

@@ -532,6 +532,9 @@ async def execute_text_chat_pending_turn(
     embeddings_api_key = None
     embeddings_model = None
     embeddings_base_url = None
+    embeddings_provider = None
+    embeddings_endpoint = None
+    embeddings_api_version = None
     if user_config.embeddings:
         from api.services.configuration.ai_model_configuration import (
             apply_managed_embeddings_base_url,
@@ -544,6 +547,8 @@ async def execute_text_chat_pending_turn(
             provider=embeddings_provider,
             base_url=getattr(user_config.embeddings, "base_url", None),
         )
+        embeddings_endpoint = getattr(user_config.embeddings, "endpoint", None)
+        embeddings_api_version = getattr(user_config.embeddings, "api_version", None)
 
     has_recordings = await db_client.has_active_recordings(workflow.organization_id)
     context_compaction_enabled = (workflow.workflow_configurations or {}).get(
@@ -560,6 +565,9 @@ async def execute_text_chat_pending_turn(
         embeddings_api_key=embeddings_api_key,
         embeddings_model=embeddings_model,
         embeddings_base_url=embeddings_base_url,
+        embeddings_provider=embeddings_provider,
+        embeddings_endpoint=embeddings_endpoint,
+        embeddings_api_version=embeddings_api_version,
         has_recordings=has_recordings,
         context_compaction_enabled=context_compaction_enabled,
     )