Move system_instruction into LLMSettings

Add `system_instruction` field to `LLMSettings` so it is runtime-updatable via settings.
For Google (GoogleLLMService, GoogleVertexLLMService), deprecate the init-time arg since it was already shipped. For Anthropic, AWS Bedrock, and OpenAI, remove the init-time arg entirely since it was never shipped.

Add system instruction prepend logic to `build_chat_completion_params` overrides in Cerebras, SambaNova, Fireworks, Mistral, and Perplexity, which build params from scratch rather than calling `super()`.

Still need to handle realtime services (OpenAI Realtime, Grok Realtime, Gemini Live).

examples/foundational/39c-multiple-mcp.py

@@ -37,7 +37,7 @@ from pipecat.processors.aggregators.llm_response_universal import (
 from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
 from pipecat.runner.types import RunnerArguments
 from pipecat.runner.utils import create_transport
-from pipecat.services.anthropic.llm import AnthropicLLMService
+from pipecat.services.anthropic.llm import AnthropicLLMService, AnthropicLLMSettings
 from pipecat.services.cartesia.tts import CartesiaTTSService, CartesiaTTSSettings
 from pipecat.services.deepgram.stt import DeepgramSTTService
 from pipecat.services.mcp_service import MCPClient
@@ -139,7 +139,12 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         Just respond with short sentences when you are carrying out tool calls.
         """
 
-        llm = AnthropicLLMService(api_key=os.getenv("ANTHROPIC_API_KEY"), system_instruction=system)
+        llm = AnthropicLLMService(
+            api_key=os.getenv("ANTHROPIC_API_KEY"),
+            settings=AnthropicLLMSettings(
+                system_instruction=system,
+            ),
+        )
 
         try:
             rijksmuseum_mcp = MCPClient(