[Improvement] consider using ColumnPosition.LAST to replace columnPosition is null

[FANNG1]

What would you like to be improved?

ColumnPosition could be first and after, null value means last in Graviton now. In the perspective of programing it's confusing.
For SparkCatalog in Iceberg, null position is not allowed.

    if (add.position() instanceof TableChange.After) {
      pendingUpdate.moveAfter(DOT.join(add.fieldNames()), referenceField);
    } else if (add.position() instanceof TableChange.First) {
      pendingUpdate.moveFirst(DOT.join(add.fieldNames()));
    } else {
      Preconditions.checkArgument(
          add.position() == null,
          "Cannot add '%s' at unknown position: %s",
          DOT.join(add.fieldNames()),
          add.position());
    }

How should we improve?

For end user, if ColumnPosition is not defined, GravitonServer should fill a default value ColumnPosition.LAST. ColumnPosition is null is illegal in the Graviton Server.

[mchades]

I hold the view that Gravitino should not fill in the null position by ColumnPosition.LAST as default.
Instead, it should be the responsibility of different catalogs to handle null values.

This is due to the different ways in which catalogs handle the null position. For example, Hive defaults to adding new columns to the end of non-partitioned columns(#871 did as the Hive default behavior), while Iceberg adds them to the last column (native Iceberg throws an exception, but #383 made an implicit conversion).

A uniform default value fill would make it difficult for catalogs to distinguish between "Last" and null.
@jerryshao What's your opinion?