patch: remove usage of llm, chat model __call__ (#20788)

- `llm(prompt)` -> `llm.invoke(prompt)`
- `llm(prompt=prompt` -> `llm.invoke(prompt)` (same with `messages=`)
- `llm(prompt, callbacks=callbacks)` -> `llm.invoke(prompt,
config={"callbacks": callbacks})`
- `llm(prompt, **kwargs)` -> `llm.invoke(prompt, **kwargs)`

libs/community/tests/unit_tests/llms/test_callbacks.py

@@ -13,7 +13,7 @@ def test_llm_with_callbacks() -> None:
     """Test LLM callbacks."""
     handler = FakeCallbackHandler()
     llm = FakeListLLM(callbacks=[handler], verbose=True, responses=["foo"])
-    output = llm("foo")
+    output = llm.invoke("foo")
     assert output == "foo"
     assert handler.starts == 1
     assert handler.ends == 1
@@ -26,7 +26,7 @@ def test_chat_model_with_v1_callbacks() -> None:
     llm = FakeListChatModel(
         callbacks=[handler], verbose=True, responses=["fake response"]
     )
-    output = llm([HumanMessage(content="foo")])
+    output = llm.invoke([HumanMessage(content="foo")])
     assert output.content == "fake response"
     assert handler.starts == 1
     assert handler.ends == 1
@@ -41,7 +41,7 @@ def test_chat_model_with_v2_callbacks() -> None:
     llm = FakeListChatModel(
         callbacks=[handler], verbose=True, responses=["fake response"]
     )
-    output = llm([HumanMessage(content="foo")])
+    output = llm.invoke([HumanMessage(content="foo")])
     assert output.content == "fake response"
     assert handler.starts == 1
     assert handler.ends == 1