[SPARK-54754][SQL] OrcSerializer should not parse the schema every time it is serialized

[leixm]

What changes were proposed in this pull request?

For maps, arrays, and structs, calling OrcUtils.orcTypeDescription(dataType) during each serialization process is quite time-consuming. We can reuse TypeDescription.

I tested calling OrcSerializer#serialize with the 10k/100k/1m rows of data, the schema of which is as follows:

private val schema: StructType = StructType(Seq(
  StructField("id", LongType, nullable = false),
  StructField("attrs",
    MapType(StringType, MapType(StringType, IntegerType, valueContainsNull = false),
      valueContainsNull = true),
    nullable = true)
))

Test result:

row count	master	pr	time saving
10k	16 ms	8 ms	50%
100k	201 ms	143 ms	29%
1m	1799 ms	1056 ms	41%

The above benchmarks were run on my local machine.

Why are the changes needed?

Improve performance of OrcSerializer#serialize.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

Existing UTs.
New benchmark (in a separate PR)

Was this patch authored or co-authored using generative AI tooling?

No.

[leixm]

ping @dongjoon-hyun Can you help review?

[yaooqinn]

+1, LGTM

[yaooqinn]

Merged to master, thank you @leixm.

BTW, user 'jerrylei' is added to Apache Spark contributor's role as JIRA side, please let me know if it's not you.