feat: Support Spark expression window_time

[0lai0]

Which issue does this PR close?

Closes #3138

Rationale for this change

Comet previously did not support the Spark PreciseTimestampConversion expression.
This expression is not called directly by users, but is generated internally by Spark's Analyzer when it rewrites window_time(window_column) into a combination of GetStructField, Subtract, and PreciseTimestampConversion. Since Comet already supports GetStructField and Subtract but not PreciseTimestampConversion, queries using window_time would fall back to JVM execution.

What changes are included in this PR?

This change adds a Serde handler for PreciseTimestampConversion that reuses the existing Cast protobuf and Rust implementation. Since TimestampType (microseconds) and LongType share the same 64-bit memory layout in Arrow, a simple cast suffices without any new native code paths.

• datetime.scala : Added CometPreciseTimestampConversion handler
• QueryPlanSerde.scala : Registered the handler in temporalExpressions map
• window_time.sql : Added window_time test

How are these changes tested?

Added a Scala unit test for window_time expression support in CometExpressionSuite

./mvnw test -Dsuites="org.apache.comet.CometExpressionSuite" -Dscala.useZincServer=false

[comphead]

Thanks @0lai0
Triggering CI, please have SQL tests in separate sql files, similar to array_append.sql

[andygrove]

checkSparkAnswer isn't checking that the query is actually running in Comet. However, it would be better to add the SQL based tests instead as @comphead pointed out

[0lai0]

Thanks @comphead and @andygrove for review and point this out. I'll add SQL tests in separate sql files and remove previous test. By the way, I discovered that using setcast is problematic, so I have updated it accordingly.