docs: sync connected components docs with the code - #903

What changes were proposed in this pull request?

The configuration page and the connected components argument list had drifted from the
implementation. Every correction below is against code on main.

docs/src/04-user-guide/13-configurations.md

Also: all six code examples set the deprecated "graphframes" value, so following the page
produced a deprecation warning. They now use "two_phase".

Two things the page never mentioned, both of which matter in practice:

• checkpointing falls back to the spark.checkpoint.dir configuration when
  SparkContext.setCheckpointDir is not set (TwoPhase.scala:278,
  RandomizedContraction.scala:69). This is the only route available when sparkContext is
  not reachable, for example from a Spark Connect client.
• two_phase in AQE mode (broadcastthreshold = -1) uses Spark's own DataFrame checkpointing
  rather than its own named sub-directory (TwoPhase.scala:534), unlike the other paths.

docs/src/04-user-guide/05-traversals.md

The connected components argument list documented checkpoint_interval,
broadcast_threshold and use_labels_as_components. Those names match neither API — the
Python parameters are checkpointInterval, broadcastThreshold and useLabelsAsComponents,
and the Scala setters are setCheckpointInterval, setBroadcastThreshold and
setUseLabelsAsComponents. Copying from the page raised TypeError. They now use the
(Python) / (Scala) convention already used further up the same page for all_paths.

maxIter is no longer described as defaulting to Integer.MAX_VALUE, since the Python and
Scala defaults differ by one; it is described as effectively unlimited instead.

One typo outside the connected components section is also fixed: the shortest paths argument
list documented is_direted; the parameter is is_directed. Happy to split that out if you
would rather keep this PR strictly scoped.

Note that checkpoint_interval is deliberately left untouched in the shortest paths, cycles
detection and aggregate neighbors sections — it is the correct Python name for those methods.
Only connectedComponents uses camelCase.

Why are the changes needed?

This is the page a user lands on when tuning connected components for a large job. As written
it gave the wrong default component ID type, hid the algorithm the project itself recommends
as roughly 5x faster, documented three parameter names that cannot be typed, and stated the
reliability of local checkpoints backwards.

Verification

Documentation only; no code, build or CI files are touched.

sbt docs/laikaHTML could not be run to completion, for a reason unrelated to this PR:
docs/generateAtomFeed fails on a clean checkout with
DateTimeParseException: Text '1970-01-01:T00:00:00Z' could not be parsed at index 10
(LaikaCustoms.scala:67 and :87 hardcode that literal, which carries a stray colon).