fix(semantic-scene): move describe_screen_semantic out of the tools/ layer

tools/ is a Hexagonal inner layer that must never depend on LLM
concerns (ADR 0002), but describe_screen_semantic.py imported
semantic.enricher, which pulls in the Anthropic client by default.
Relocated the wrapper to runtime/, which is where LLM-dependent code
is allowed to live; updated the tool registry and all test imports
accordingly. No behavior change.

openspec: semantic-scene capability, archived change semantic-scene-runtime

tests/test_semantic_task_loop.py

@@ -6,12 +6,12 @@ import pytest
 
 from core.models import Bounds, Scene, SceneElement, Task
 from device.manager import DeviceManager
+from runtime.describe_screen_semantic import describe_screen_semantic
 from runtime.executor import Executor, ExecutorConfig
 from runtime.planner import PlannedStep, Planner
 from runtime.task import TaskRunner, TaskRunnerConfig
 from semantic.config import SemanticConfig
 from semantic.llm_client import EnrichmentUnavailable
-from tools.describe_screen_semantic import describe_screen_semantic
 from tests.fakes import FakeDriver, PNG_10X20