Interval Literal doesn't work for timeunit less than millisecond

[waitingkuo]

Describe the bug
A clear and concise description of what the bug is.

To Reproduce
Steps to reproduce the behavior:

It works for millisecond

❯ select interval '1 month 0.001 second';
+-------------------------------------------------------+
| IntervalMonthDayNano("79228162514264337593544950336") |
+-------------------------------------------------------+
| 0 years 1 mons 0 days 0 hours 0 mins 0.1000000 secs   |
+-------------------------------------------------------+
1 row in set. Query took 0.001 seconds.

but doesn't work while the unit for second less than millisecond

❯ select interval '1 month 0.0001 second';
+------------------------------------------------+
| IntervalYearMonth("1")                         |
+------------------------------------------------+
| 0 years 1 mons 0 days 0 hours 0 mins 0.00 secs |
+------------------------------------------------+
1 row in set. Query took 0.002 seconds.

Expected behavior
A clear and concise description of what you expected to happen.

Additional context
Add any other context about the problem here.

[Jefffrey]

Seems to be because of this parsing:

https://github.com/apache/arrow-datafusion/blob/129654cd8466e7634bf186782c3d271541f0671e/datafusion/common/src/parsers.rs#L130-L161

Finest granularity is milliseconds, accumulated as an integer, hence fractions of that are silently ignored.

Could change this accumulator to nanoseconds, as i64, however this seems it would conflict with ScalarValue::IntervalDayTime which stores milliseconds as i32, compared to ScalarValue::IntervalMonthDayNano which does store nanoseconds.

Might need to change ScalarValue::IntervalDayTime to support nanoseconds perhaps? If want to be consistent between precision of ScalarValue::IntervalMonthDayNano and ScalarValue::IntervalDayTime

Edit: though since IntervalDayTime is an Arrow type, maybe just have that if fractions of milliseconds is found, then the result should always be a IntervalMonthDayNano to support keeping this precision, and thus not forcing IntervalDayTime to silently drop those fractional parts?