[FEA] New kernel to support parsing dates/timestamps string with a timezone parameter.

[res-life]

Is your feature request related to a problem? Please describe.
ToUnixTimestamp and GetTimestamp or more operators require a format parameter, and they are time zone aware operators. Refer to Spark link

val formatter = formatterOption.getOrElse(getFormatter(fmt.toString))
formatter.parse()

I think our GPU implemetation currently does not support non-utc TZ:

def parseStringAsTimestamp(

When TZ is Asia/Shanghai, to_timestamp("1970-01-01 00:00:00", "yyyy-MM-dd HH:mm:ss") get negative 8 hours instead of zero.

Describe the solution you'd like
Expose the timeparts structure, refer to the following code.
Then rebase local time in a time zone to UTC time. Alfred is woring on this.

https://github.com/rapidsai/cudf/blob/v24.02.00a/cpp/src/strings/convert/convert_datetime.cu#L399-L401

    auto const timeparts = parse_into_parts(d_str);

    return T{T::duration(timestamp_from_parts(timeparts))};

• We can update cudf code.
• We can copy cudf code. I prefer this option.
  We have limited supported format

yyyy-MM-dd
yyyy/MM/dd
yyyy-MM
yyyy/MM
dd/MM/yyyy
yyyy-MM-dd HH:mm:ss
MM-dd
MM/dd
dd-MM
dd/MM
MM/yyyy
MM-yyyy
MM/dd/yyyy
MM-dd-yyyy
MMyyyy

We can add more supported format in the future.

• Or we can put the code into this PR: Add timestamp parser to parse timestamp string with time zone #1539

• legacy format: link java.text.SimpleDateFormat

• new format: java.time.format.DateTimeFormatter

• When spark.sql.legacy.timeParserPolicy is EXCEPTION, refer to link

[res-life]

As Haoyang said, we can handle this issue together: NVIDIA/spark-rapids#10032

[res-life]

Update:
After sync-up with @NVnavkumar.
Maybe we have a simple solution via GpuTimeZoneDB directly.
We first get the microseconds from 1970-01-01 00:00:00 in UTC which are called instants.
Then use GpuTimeZoneDB to rebase the microseconds according to the timezone.

For example: Parse("1970-01-01 00:00:00", 'yyyy-MM-dd HH:mm:ss') when session time zone is Aisa/Shanghai.

• First get the microseconds, here it's 0. This means get microseconds from "1970-01-01 00:00:00" in UTC TZ.
• Then use GpuTimeZoneDB.fromUTC(cv, tz) or GpuTimeZoneDB.toUTC(cv, tz) to rebase the microseconds to what we want.

I think NVIDIA/spark-rapids#10100 will close this, only need to double check.
@thirtiseven If above works, then we only need to handle: NVIDIA/spark-rapids#10032

[thirtiseven]

The simple solution works, close it and let 10032 track the other parts.