[SPARK-56663][SQL][FOLLOWUP] Fix silent overflow in date_trunc fast path.

### What changes were proposed in this pull request?

Fixes the bug introduced by #55610. When `local` is a very large negative value, `Math.floorMod(local, unitMicros)` is positive because `unitMicros` is positive, and the subtraction can overflow to positive values. This produces a inconsistent result from the slow path.

In most common cases, this optimization still takes effect. The performance loss is ignorable.

### Why are the changes needed?

Fix incorrect result.

### Does this PR introduce _any_ user-facing change?

No.

### How was this patch tested?

New unit tests. It passes before #55610, or after this change. But it fails on the current master.

### Was this patch authored or co-authored using generative AI tooling?

No.

Closes #56313 from chenhao-db/fix_trunc_overflow.

Authored-by: chenhao-db <chenhao.li@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
(cherry picked from commit 94d23c3)
Signed-off-by: Wenchen Fan <wenchen@databricks.com>

Author: chenhao-db

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/util/DateTimeUtils.scala

@@ -510,7 +510,7 @@ object DateTimeUtils extends SparkDateTimeUtils {
     val offsetMicros = originalOffsetSec * MICROS_PER_SECOND
     try {
       val local = Math.addExact(micros, offsetMicros)
-      val truncatedLocal = local - Math.floorMod(local, unitMicros)
+      val truncatedLocal = Math.subtractExact(local, Math.floorMod(local, unitMicros))
       val candidate = Math.subtractExact(truncatedLocal, offsetMicros)
       if (!rules.isFixedOffset) {
         val candidateSec = Math.floorDiv(candidate, MICROS_PER_SECOND)
@@ -537,9 +537,9 @@ object DateTimeUtils extends SparkDateTimeUtils {
     level match {
       case TRUNC_TO_MICROSECOND => micros
       case TRUNC_TO_MILLISECOND =>
-        micros - Math.floorMod(micros, MICROS_PER_MILLIS)
+        Math.subtractExact(micros, Math.floorMod(micros, MICROS_PER_MILLIS))
       case TRUNC_TO_SECOND =>
-        micros - Math.floorMod(micros, MICROS_PER_SECOND)
+        Math.subtractExact(micros, Math.floorMod(micros, MICROS_PER_SECOND))
       case TRUNC_TO_MINUTE =>
         truncToUnitFast(micros, zoneId, MICROS_PER_MINUTE, ChronoUnit.MINUTES)
       case TRUNC_TO_HOUR =>

sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/DateExpressionsSuite.scala

@@ -853,6 +853,21 @@ class DateExpressionsSuite extends SparkFunSuite with ExpressionEvalHelper {
     }
   }
 
+  test("TruncTimestamp of Long.MinValue overflows with ArithmeticException") {
+    withDefaultTimeZone(UTC) {
+      // Long.MinValue is the smallest representable timestamp value (in micros). Truncating it
+      // rounds the value down to an earlier instant, which falls below the representable micros
+      // range. The overflow must surface as an ArithmeticException instead of silently wrapping
+      // around to a bogus (positive) timestamp.
+      val minTimestamp = Literal.create(Long.MinValue, TimestampType)
+      Seq("YEAR", "QUARTER", "MONTH", "WEEK", "DAY", "HOUR", "MINUTE",
+          "SECOND", "MILLISECOND").foreach { fmt =>
+        checkExceptionInExpression[ArithmeticException](
+          TruncTimestamp(Literal.create(fmt, StringType), minTimestamp), "")
+      }
+    }
+  }
+
   test("unsupported fmt fields for trunc/date_trunc results null") {
     Seq("INVALID", "decade", "century", "millennium", "whatever", null).foreach { field =>
       testTruncDate(Date.valueOf("2000-03-08"), field, null)