[WIP][SPARK-44243][CORE] Add a parameter to determine the locality of local shuffle reader

[wankunde]

What changes were proposed in this pull request?

Follow changes of #40339

Local shuffle reader can achieve better performance with preferred locations. If we disable SHUFFLE_REDUCE_LOCALITY_ENABLE in queries that include reduce shuffles and local shuffles, local shuffle readers can not get preferred locations.

Add new parameter LOCAL_SHUFFLE_LOCALITY_ENABLE to determine whether to get the preferred locations of the current partitionSpec.

Why are the changes needed?

Improvement for spark local shuffle.

Does this PR introduce any user-facing change?

No

How was this patch tested?

Exists UT

[mridulm]

If you want locality for shuffle, enable spark.shuffle.reduceLocality.enabled - why introduce another config for this ?

[HyukjinKwon]

cc @maryannxue FYI

[maryannxue]

What's the use of the new conf? How does it improve locality? Isn't it just enough to do https://github.com/apache/spark/pull/41786/files#diff-a3b15298f97577c1fadcc2d76d015eebd6343e246c6717417d33f3c458847f46L1133?

[wankunde]

Thanks @mridulm @maryannxue for your review.

If a query contains shuffle A and shuffle B, there are many PartialReducerPartitions after OptimizeSkewedJoin optimization, and shuffle B is a local read shuffle. Enable spark.shuffle.reduceLocality.enabled may takes some extra time to get the preferred locations.
But there is a limit on the number of PartialReducerPartitions in our production environment, so it is okay for me not to make this change.