Refactor workflow agent and routing components for improved functionality

- Introduce WorkflowAgentStage to manage agent stage configurations and enhance interaction with the workflow engine.
- Implement WorkflowEdgeEvaluator for priority-aware edge evaluation, improving routing decisions based on conditions and user turns.
- Update WorkflowBrain to handle user turns and routing more effectively, ensuring agents cannot have only one default path.
- Enhance CallEndCoordinator to track speech events and manage call termination based on queued speech.
- Add new models and output handling for workflow interactions, improving clarity and maintainability.
- Update tests to validate the new routing logic and agent behavior under various scenarios.

backend/services/workflow/models.py

@@ -0,0 +1,92 @@
+"""Readable runtime values shared by Workflow orchestration modules."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from enum import StrEnum
+from typing import Any
+
+
+class WorkflowStatus(StrEnum):
+    """The small set of states useful to operators and future debug tooling."""
+
+    STARTING = "starting"
+    WAITING_USER = "waiting_user"
+    ROUTING = "routing"
+    RUNNING_AGENT = "running_agent"
+    RUNNING_ACTION = "running_action"
+    HANDOFF = "handoff"
+    ENDED = "ended"
+
+
+class RouteStatus(StrEnum):
+    """A routing error is deliberately different from a valid no-match."""
+
+    MATCHED = "matched"
+    NO_MATCH = "no_match"
+    ERROR = "error"
+
+
+@dataclass(frozen=True)
+class UserTurn:
+    """One committed user turn that may cross automatic Workflow nodes."""
+
+    id: int
+    text: str
+
+
+@dataclass
+class WorkflowRuntimeState:
+    """Mutable per-call state; graph definitions remain immutable."""
+
+    current_node_id: str
+    status: WorkflowStatus = WorkflowStatus.STARTING
+    pending_user_turn: UserTurn | None = None
+    transition_id: int = 0
+    automatic_hops: int = 0
+    ended: bool = False
+    _next_turn_id: int = 1
+
+    def begin_user_turn(self, text: str) -> UserTurn:
+        turn = UserTurn(id=self._next_turn_id, text=text)
+        self._next_turn_id += 1
+        self.pending_user_turn = turn
+        self.automatic_hops = 0
+        return turn
+
+    def enter(self, node_id: str, status: WorkflowStatus) -> None:
+        self.current_node_id = node_id
+        self.status = status
+
+    def begin_transition(self) -> int:
+        self.transition_id += 1
+        return self.transition_id
+
+    def consume_user_turn(self) -> UserTurn | None:
+        turn = self.pending_user_turn
+        self.pending_user_turn = None
+        return turn
+
+    def finish(self) -> None:
+        self.ended = True
+        self.status = WorkflowStatus.ENDED
+        self.pending_user_turn = None
+
+
+@dataclass(frozen=True)
+class LLMRouteResult:
+    """Control-plane result returned by the small routing LLM."""
+
+    status: RouteStatus
+    function_name: str | None = None
+    error: str | None = None
+
+
+@dataclass(frozen=True)
+class EdgeEvaluation:
+    """Final graph-level decision after expression, LLM and default handling."""
+
+    status: RouteStatus
+    edge: dict[str, Any] | None = None
+    error: str | None = None
+