[BUG] DateType writes use JVM default timezone midnight, shifting the calendar day on non UTC JVMs

[lawofcycles]

What is the bug?

When the Spark connector writes a Spark DateType value to OpenSearch, DataFrameValueWriter serializes it to epoch millis via value.asInstanceOf[Date].getTime(). A java.sql.Date represents midnight in the JVM default timezone, so on a non UTC JVM this writes local midnight instead of UTC midnight and the stored value can land on a different calendar day. The same code is present in the sql-30, sql-35, and sql-40 modules.

How can one reproduce the bug?

1. Set the JVM default timezone to a non UTC zone such as Asia/Tokyo (UTC+9).
2. Create a DataFrame with a DateType column holding 2023-07-22.
3. Write it to an OpenSearch index whose field is a date.
4. Observe the stored value is 2023-07-21T15:00:00Z, so it reads back as the previous day.

What is the expected behavior?

A DateType is a timezone independent calendar date, so it should be stored as UTC start of day (2023-07-22T00:00:00Z). A tz independent conversion such as d.toLocalDate.atStartOfDay(java.time.ZoneOffset.UTC).toInstant.toEpochMilli produces the correct value.

What is your host/environment?

Any non UTC JVM timezone, for example Asia/Tokyo. Reproducible on current main across all Spark modules.

Do you have any screenshots?

NA

Do you have any additional context?

NA