Spark: MERGE INTO Statements with only WHEN NOT MATCHED Clauses are always executed at Snapshot Isolation

### Apache Iceberg version

1.8.1 (latest release)

### Query engine

Spark

### Please describe the bug 🐞

When running two concurrent `MERGE INTO` operations on an Apache Iceberg table, I expect them to be **idempotent** -- meaning Iceberg should either detect conflicts and resolve them or fail one of the jobs to prevent data inconsistencies.

However, Iceberg determines the **operation type dynamically** based on the result of the join condition, which can lead to unexpected behavior:
- If a match is found, Iceberg treats it as an **overwrite** operation and fails the second job due to conflicting commits.
- If no match is found, Iceberg considers it an **append** operation and attempts to resolve conflicts by creating a new manifest for appended data, as explained in the [Cost of Retries doc](https://iceberg.apache.org/docs/latest/reliability/#cost-of-retries).

This behavior introduces a problem:
If the dataset is large enough and neither job finds a match, both will proceed with appending data independently, causing **duplicate records**.

#### **Reproduction Steps**
Running the following query in concurrent jobs can result in duplicate data if no matching records exist in `dest`:
```SQL
MERGE INTO dest
USING src
ON dest.id = src.id
WHEN NOT MATCHED THEN
INSERT *
-- even with update action, we'll have the same issue
-- WHEN MATCHED THEN
-- UPDATE SET *
```
I initially expected the **operation type** to be determined by the query itself (i.e., always "append" in the query without `UPDATE` action). However, through testing, I found that Iceberg decides the operation type **at runtime**, based on the actual join results. This makes `MERGE INTO` **non-idempotent**, leading to unintended duplicate inserts.

#### **Expected Behavior**
Iceberg should ensure idempotency for `MERGE INTO`, preventing duplicate data when no matches are found.

#### **Additional Context**
- Iceberg version: 1.8.1
- Iceberg catalog: Glue catalog (type `glue`) with S3 FileIO
- Spark version: 3.5.5

Would love to hear if others have encountered this or if there's a recommended workaround.

### Willingness to contribute

- [ ] I can contribute a fix for this bug independently
- [x] I would be willing to contribute a fix for this bug with guidance from the Iceberg community
- [ ] I cannot contribute a fix for this bug at this time

Commencez par reproduire les opérations Spark MERGE INTO concurrentes sur une table Iceberg décrites dans l’issue, en utilisant la requête qui ne contient que WHEN NOT MATCHED et les versions de Spark et d’Iceberg indiquées. Lisez la documentation Cost of Retries et suivez la manière dont le résultat de la jointure à l’exécution sélectionne le comportement d’append ou d’overwrite ; le travail est terminé lorsque les merges concurrents sans correspondance ne créent pas de doublons et que le comportement est couvert par des tests de régression.

Rédigé par le modèle d'indexation à partir du texte de l'issue.