GeezerVison

What changes were proposed in this pull request?

Corrects the docs on useLocalCheckpoints for randomized_contraction. They currently say local checkpoints remove the need for a checkpoint directory for both connected components algorithms, but that holds for two_phase only.

Changed: docs/src/04-user-guide/13-configurations.md (four sentences) and docs/src/04-user-guide/05-traversals.md (one sentence added to use_local_checkpoints). No code changes.

Why are the changes needed?

Following the current docs, a user who sets use_local_checkpoints=True for randomized_contraction still gets java.io.IOException: Checkpoint directory is not set:

g = GraphFrame(
    spark.createDataFrame([(1,), (2,), (3,)], ["id"]),
    spark.createDataFrame([(1, 2)], ["src", "dst"]),
)

# works
g.connectedComponents(algorithm="two_phase", use_local_checkpoints=True).collect()

# throws
g.connectedComponents(algorithm="randomized_contraction", use_local_checkpoints=True).collect()

On Spark Connect this leaves no way forward from the client, since spark.checkpoint.dir is a static conf there (CANNOT_MODIFY_CONFIG).