[Fix] streaming chatbedrock converseapi (#351)

Added run_manager in stream method of ChatBedrockConverse. Added
streaming flag handling for to enable converse streaming with
ChatBedrock
 
 Fixes #334 

### Using `Chatbedrock` with `streaming` flag set `True`:

```python
import boto3
from langchain_aws import ChatBedrock, ChatBedrockConverse
from langchain.callbacks.base import BaseCallbackHandler
from langchain_core.prompts import ChatPromptTemplate

streaming = True
session = boto3.session.Session()
bedrock_client = session.client("bedrock-runtime", region_name="us-east-1")

class MyCustomHandler(BaseCallbackHandler):
    def on_llm_new_token(self, token: str, **kwargs) -> None:
        print(f"My custom handler, token: {token}; Type of token: {type(token)}")

prompt = ChatPromptTemplate.from_messages(["Tell me a joke about {animal}"])

model = ChatBedrock(
    client=bedrock_client, 
    model_id="anthropic.claude-3-haiku-20240307-v1:0",
    region_name="us-east-1",
    streaming = streaming,
    callbacks=[MyCustomHandler()],
)

chain = prompt | model

response = chain.invoke({"animal": "bears"}) 
print(response)
```


### Output:

```
(base) vishal@mypc aws % poetry run python3 streaming_demo.py
My custom handler, token: ; Type of token: <class 'str'>
My custom handler, token: Here; Type of token: <class 'str'>
My custom handler, token: 's; Type of token: <class 'str'>
My custom handler, token:  a; Type of token: <class 'str'>
My custom handler, token:  bear; Type of token: <class 'str'>
My custom handler, token:  joke; Type of token: <class 'str'>
My custom handler, token:  for; Type of token: <class 'str'>
My custom handler, token:  you; Type of token: <class 'str'>
My custom handler, token: :; Type of token: <class 'str'>
My custom handler, token: 
...
...
...
```


### Using `Chatbedrock` with `streaming` and `beta_use_converse_api`
flags set as `True`:

```python
import boto3
from langchain_aws import ChatBedrock, ChatBedrockConverse
from langchain.callbacks.base import BaseCallbackHandler
from langchain_core.prompts import ChatPromptTemplate

streaming = True
session = boto3.session.Session()
bedrock_client = session.client("bedrock-runtime", region_name="us-east-1")

class MyCustomHandler(BaseCallbackHandler):
    def on_llm_new_token(self, token: str, **kwargs) -> None:
        print(f"My custom handler, token: {token}; Type of token: {type(token)}")

prompt = ChatPromptTemplate.from_messages(["Tell me a joke about {animal}"])

model = ChatBedrock(
    client=bedrock_client, 
    model_id="anthropic.claude-3-haiku-20240307-v1:0",
    region_name="us-east-1",
    streaming = streaming,
    callbacks=[MyCustomHandler()],
    beta_use_converse_api = True
)

chain = prompt | model

response = chain.invoke({"animal": "bears"}) 
print(response)
```

### Output:

```
(base) vishal@mypc aws % poetry run python3 streaming_demo.py
My custom handler, token: ; Type of token: <class 'str'>
My custom handler, token: Here; Type of token: <class 'str'>
My custom handler, token: 's; Type of token: <class 'str'>
My custom handler, token:  a; Type of token: <class 'str'>
My custom handler, token:  bear; Type of token: <class 'str'>
My custom handler, token:  joke; Type of token: <class 'str'>
My custom handler, token:  for; Type of token: <class 'str'>
My custom handler, token:  you; Type of token: <class 'str'>
My custom handler, token: :; Type of token: <class 'str'>
My custom handler, token: 
...
...
...

```



#### Stilling figuring out the following:

Using `ChatBedrock` with `stream()` method and the custom handler gives
similar output. But the same with `ChatBedrockConverse` gives a
different output as shown below for the given code:

```python
import boto3
from langchain_aws import ChatBedrock, ChatBedrockConverse
from langchain.callbacks.base import BaseCallbackHandler
from langchain_core.prompts import ChatPromptTemplate

streaming = True
session = boto3.session.Session()
bedrock_client = session.client("bedrock-runtime", region_name="us-east-1")

class MyCustomHandler(BaseCallbackHandler):
    def on_llm_new_token(self, token: str, **kwargs) -> None:
        print(f"My custom handler, token: {token}; Type of token: {type(token)}")

prompt = ChatPromptTemplate.from_messages(["Tell me a joke about {animal}"])

model = ChatBedrock(
    client=bedrock_client, 
    model_id="anthropic.claude-3-haiku-20240307-v1:0",
    region_name="us-east-1",
    callbacks=[MyCustomHandler()],
    beta_use_converse_api = True
)

chain = prompt | model

for chunk in chain.stream({"animal": "bears"}):
    pass
```

### Output:

```
My custom handler, token: []; Type of token: <class 'list'>
My custom handler, token: [{'type': 'text', 'text': 'Here', 'index': 0}]; Type of token: <class 'list'>
My custom handler, token: [{'type': 'text', 'text': "'s", 'index': 0}]; Type of token: <class 'list'>
My custom handler, token: [{'type': 'text', 'text': ' a', 'index': 0}]; Type of token: <class 'list'>
My custom handler, token: [{'type': 'text', 'text': ' bear', 'index': 0}]; Type of token: <class 'list'>
My custom handler, token: [{'type': 'text', 'text': '-', 'index': 0}]; Type of token: <class 'list'>
My custom handler, token: [{'type': 'text', 'text': 'y', 'index': 0}]; Type of token: <class 'list'>
My custom handler, token: [{'type': 'text', 'text': ' goo', 'index': 0}]; Type of token: <class 'list'>
My custom handler, token: [{'type': 'text', 'text': 'd joke', 'index': 0}]; Type of token: <class 'list'>
My custom handler, token: [{'type': 'text', 'text': ' for', 'index': 0}]; Type of token: <class 'list'>
...
...
...

```

---------

Co-authored-by: Vishal Patil <[email protected]>
Co-authored-by: Michael Chin <[email protected]>

libs/aws/langchain_aws/chat_models/bedrock.py

@@ -526,10 +526,15 @@ def _generate(
         **kwargs: Any,
     ) -> ChatResult:
         if self.beta_use_converse_api:
-            return self._as_converse._generate(
-                messages, stop=stop, run_manager=run_manager, **kwargs
-            )
-
+            if not self.streaming:
+                return self._as_converse._generate(
+                    messages, stop=stop, run_manager=run_manager, **kwargs
+                )
+            else:
+                stream_iter = self._as_converse._stream(
+                    messages, stop=stop, run_manager=run_manager, **kwargs
+                )
+                return generate_from_stream(stream_iter)
         completion = ""
         llm_output: Dict[str, Any] = {}
         tool_calls: List[ToolCall] = []

libs/aws/langchain_aws/chat_models/bedrock_converse.py

@@ -542,7 +542,12 @@ def _stream(
         )
         for event in response["stream"]:
             if message_chunk := _parse_stream_event(event):
-                yield ChatGenerationChunk(message=message_chunk)
+                generation_chunk = ChatGenerationChunk(message=message_chunk)
+                if run_manager:
+                    run_manager.on_llm_new_token(
+                        generation_chunk.text, chunk=generation_chunk
+                    )
+                yield generation_chunk
 
     # TODO: Add async support once there are async bedrock.converse methods.