Prune full-data-file Iceberg changelog delete tasks by metadata predicates

**Is your feature request related to a problem? Please describe.**

Auron already supports native execution of `DeletedDataFileScanTask` when `existingDeletes()` is empty. However, `filterChangelogTasks` evaluates metadata predicates only for `AddedRowsScanTask` and retains all other tasks.

As a result, a supported full-data-file DELETE task remains in the native scan plan even when its metadata proves that it cannot match the query. See the [current task filtering implementation](https://github.com/apache/auron/blob/04eeb90d2911465eb350253eacc281a16bb5f37f/thirdparty/auron-iceberg/src/main/scala/org/apache/spark/sql/auron/iceberg/IcebergScanSupport.scala#L614-L634).

For example, given a changelog range containing only full-data-file DELETE tasks:

```sql
SELECT id
FROM changes
WHERE _change_type = 'INSERT';
```

The result is empty, but the DELETE tasks are retained. The existing integration test explicitly checks this behavior and asserts `numFiles == 1` after executing the query. See the [existing regression test](https://github.com/apache/auron/blob/04eeb90d2911465eb350253eacc281a16bb5f37f/thirdparty/auron-iceberg/src/test/scala/org/apache/auron/iceberg/AuronIcebergIntegrationSuite.scala#L810-L821).

The same limitation applies when `_commit_snapshot_id` or `_change_ordinal` excludes a DELETE task. This leaves unnecessary scan work in metadata-filtered changelog queries.

**Describe the solution you'd like**

Extend the existing changelog metadata pruning to supported `DeletedDataFileScanTask` instances with empty `existingDeletes()`.

Reuse the current per-task metadata values and predicate evaluation for:

- `_change_type`
- `_commit_snapshot_id`
- `_change_ordinal`

Support the existing `=` and `IN` predicates combined with `AND`, and retain the complete post-scan filter.

Apply pruning after the existing native task eligibility checks. Preserve the expression-ID checks, operator boundaries, runtime-filter tag propagation, and fallback behavior for unsupported tasks.

This change should not introduce position/equality delete handling or extend predicate extraction to `OR`, `NOT`, or mixed metadata/data-column conditions.

Validation should cover:

- A DELETE-only range filtered by `_change_type = 'INSERT'`: empty results and zero retained native scan tasks.
- A matching DELETE predicate: matching tasks and rows remain.
- DELETE tasks from different snapshots: equality, `IN`, and `AND` predicates on snapshot IDs and change ordinals retain only matching tasks.
- A range containing both INSERT and full-data-file DELETE tasks: either task type can be excluded independently.
- A predicate that excludes every task: execution succeeds with empty results.
- Unsupported predicates and tasks: existing behavior is preserved.

Check explicit expected rows and compare Auron-enabled execution with Spark. For eligible cases, assert native scan conversion and inspect `scanTasks` to verify pruning; result equality alone does not demonstrate that scan work was reduced.

**Describe alternatives you've considered**

Continue relying on the post-scan filter. This preserves query results, but retains DELETE tasks that can already be excluded using task-level metadata.

**Additional context**

This is a follow-up to:

- [#2435](https://github.com/apache/auron/pull/2435), which added native full-data-file DELETE changelog scans.
- [#2456](https://github.com/apache/auron/pull/2456), which added metadata pruning for INSERT tasks and explicitly retained non-`AddedRowsScanTask` tasks.

The expected implementation scope is `IcebergScanSupport.scala` and focused regression coverage in `AuronIcebergIntegrationSuite.scala`. No new configuration or public API is needed.

This proposal is based on source inspection and existing test assertions; no new reproduction or performance benchmark has been run.

Start in thirdparty/auron-iceberg/src/main/scala/org/apache/spark/sql/auron/iceberg/IcebergScanSupport.scala, reading filterChangelogTasks and its existing native task eligibility checks. Run the focused cases in thirdparty/auron-iceberg/src/test/scala/org/apache/auron/iceberg/AuronIcebergIntegrationSuite.scala around the existing regression test. Done means supported full-data-file DELETE tasks are pruned by the listed metadata predicates, while results, fallback behavior, and unsupported cases remain unchanged.

Written by the indexing model from the issue text.