Python: support checkpointing at runtime and not as workflow attribute

[moonbox3]

Today, when checkpointing is used, it must be enabled via the workflow builder. This doesn't allow us to support all types of checkpointing scenarios gracefully. We should allow checkpointing to be passed in to a workflow at runtime - if the checkpoint manager is passed in, then we use it. Other workflows can choose to not include checkpointing by not including the checkpoint manager.

[TaoChenOSU]

This may coincide with the API unification suggestion that @victordibia brought up. Instead of setting up the checkpoint manager in the workflow builder, we could have the following APIs for invoking workflows:

async def run_stream(self, message: Any, checkpoint: WorkflowCheckpoint | None = None, checkpoint_storage: CheckpointStorage | None = None):
    # Resume from a checkpoint if `checkpoint` is created. Create new checkpoints if `checkpoint_storage` is provided.
    # `checkpoint` won't be modified

async def run(self, message: Any, checkpoint: WorkflowCheckpoint | None = None, checkpoint_storage: CheckpointStorage | None = None):

And then we can remove run_stream_from_checkpoint and run_from_checkpoint.

This could be the first step to reducing our public API surface.

[moonbox3]

Related to #1536