[SPARK-29524][SQL] Support unordered interval units in casting from strings

[MaxGekk]

What changes were proposed in this pull request?

In the PR, I propose to replace existing implementation of CalendarInterval.fromCaseInsensitiveString based on a regexp by another implementation using finite state machine. The existing regex assumes particular order of interval units from YEAR to MICROSECOND. Modification of the regexp becomes very hard to support new features.

Why are the changes needed?

• This improves Spark SQL UX by allowing users to specify interval units in any order
• Existing regex is hard to extend
• To maintain feature parity with PostgreSQL

Does this PR introduce any user-facing change?

No

How was this patch tested?

• By existing tests in CalendarIntervalSuite
• Add new test for invalid unit dday
• By new tests for uniqueness and unordered interval units in CalendarIntervalSuite

[SparkQA]

Test build #112339 has finished for PR 26180 at commit b054346.

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

[MaxGekk]

@cloud-fan Please, take a look at it.

[cloud-fan]

Let's wait for #26190 first. I'm not sure it's the right direction to keep improving the hand-written regex-based parser instead of using antlr.

[MaxGekk]

@cloud-fan WDYT about modifying it to support UTF8String as the input and use UTF8String methods inside. This can be similar to what we have for timestamps and dates - DateTimeUtils.stringToTimestamp and stringToDate. And this should be faster, and could be used from the Cast expression.

[cloud-fan]

that sounds good!

[MaxGekk]

ok. I will close this PR, and will try to implement interval parsing from UTF8String in the future.

[MaxGekk]

@cloud-fan Here is the PR #26256 for UTF8String to CalendarInterval.