[SPARK-31762][SQL][FOLLOWUP] Avoid double formatting in legacy fractional formatter

[MaxGekk]

What changes were proposed in this pull request?

Currently, the legacy fractional formatter is based on the implementation from Spark 2.4 which formats the input timestamp twice:

    val timestampString = ts.toString
    val formatted = legacyFormatter.format(ts)

to strip trailing zeros. This PR proposes to avoid the first formatting by forming the second fraction directly.

Why are the changes needed?

It makes legacy fractional formatter faster.

Does this PR introduce any user-facing change?

No

How was this patch tested?

By existing test "format fraction of second" in TimestampFormatterSuite + added test for timestamps before 1970-01-01 00:00:00Z

[SparkQA]

Test build #123120 has finished for PR 28643 at commit 046ea8a.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #123135 has finished for PR 28643 at commit 3e0b9a4.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[MaxGekk]

I borrowed the implementation from JDK 11 java.sql.Timestamp.toString:
https://github.com/AdoptOpenJDK/openjdk-jdk11/blob/master/src/java.sql/share/classes/java/sql/Timestamp.java#L266-L313

[cloud-fan]

Does perf matter so much here? It's much simpler to write

while (nanos % 10 == 0) {
  nanos /= 10
}
formatted + "." + nanos.toString

[MaxGekk]

This code is wrong. For example, if nanos = 000001000. You removed trailing zeros in the loop, and get just 1. So, the result will be .1 which is wrong. It should be .000001.

[MaxGekk]

@cloud-fan @HyukjinKwon Please, review this PR.

[SparkQA]

Test build #123189 has finished for PR 28643 at commit eeff55a.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #123191 has finished for PR 28643 at commit 292311a.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[cloud-fan]

thanks, merging to master!

It's not a perf regression so I didn't backport it to 3.0