[VL] Spark 4.x: JNI and Velox exception handling loses Spark error condition and exception type

## Backend
VL (Velox)

## Bug description

Gluten's exception handling has two layers of information loss that cause ~49 test excludes and 2 disabled suites:

**Sub-problem 1: JNI → Java exception rewriting**
Java-side exceptions are replaced with `GlutenException` when crossing the JNI boundary, losing the original exception type and error condition code. Related: #11825.

**Sub-problem 2: Velox C++ → Spark exception mapping**
Velox errors go through the `JNI_METHOD_END` macro and are uniformly thrown as `GlutenException`. Spark expects specific subtypes like `SparkRuntimeException` with error condition codes (e.g., `NOT_NULL_ASSERT_VIOLATION`).

Parent issue: #11550

### Impact

~49 excludes + 2 TODO suites:

| Category | Excludes | Affected suites |
|----------|:--------:|-----------------|
| assert_not_null (NOT_NULL_ASSERT_VIOLATION) | 14 | RuntimeNullChecksV2Writes(7), InsertSuite(1), DataSourceV2SQL(1), Delta/GroupBased(5) |
| merge cardinality (prefix excludes) | 24 (18 real + 6 collateral) | DeltaBasedMerge(x2), GroupBasedMerge |
| raise_error / assert_true | 3 | ColumnExpressionSuite(2), CachedTableSuite(1) |
| SparkRuntimeException → SparkException | 2 | DataFrameFunctionsSuite |
| Exception class/message mismatch | 4 | JsonExpressionsSuite(1), SQLQuerySuite(3) |
| TODO suites | 2 | GlutenGroupBasedUpdateTableSuite(1), GlutenSimpleSQLViewSuite(1) |

### Cross-reference

GlutenSimpleSQLViewSuite has 2 failures: 1 belongs to this issue (error condition), 1 belongs to the decimal precision issue (SPARK-53968, spark41-only).

Start at the JNI_METHOD_END macro and read the related JNI exception handling from #11825, using parent issue #11550 for context. Compare behavior against the listed Spark suites, including RuntimeNullChecksV2Writes, merge suites, and GlutenSimpleSQLViewSuite; done means the original exception types and error condition codes are preserved and the listed excludes and TODOs are resolved.

Written by the indexing model from the issue text.