DOC: Langgraph checkpoint postgres setup is incomplete

[alberto-agudo]

Issue with current documentation:

I have recently started using the langgraph postgres checkpointer.

When reviewing the documentation, I've noticed the following:

langgraph/libs/checkpoint-postgres/README.md

Line 15 in 3a74514

> When manually creating Postgres connections and passing them to `PostgresSaver` or `AsyncPostgresSaver`, make sure to include `autocommit=True` and `row_factory=dict_row` (`from psycopg.rows import dict_row`). See a full example in this [how-to guide](https://langchain-ai.github.io/langgraph/how-tos/persistence_postgres/).

I understand why the autocommit=True parameter is currently requested from a pre-configured connection provided for the PostgresSaver class as it comes handy and avoid problems with the .setup() method not committing the checkpoint tables to the DB. However, row_factory=dict_row is also "necessary" according to the README, but there should be more detail into why this is necessary.
I've set up my current project with the default tuple_row() and it's working properly, but changing this would not be backwards-compatible right now. Hence, I'd need more justification to see why I'd need to switch to this setting. I think it would generally benefit other developers.

Idea or request for content:

Detail in the PostgresSaver class or elsewhere in the repo why do we need to use autocommit=True and row_factory=dict_row