ORC: don't push file filters for omitted initial-default fields

[cbb330]

Summary

When a defaulted column is omitted from the ORC file projection (#267), Spark may still filter on it (WHERE country = 'US'). Pushing that filter into the file fails in two ways:

1. If the projection is only that omitted column, the read schema is empty and converting the filter crashes.
2. If the filter is pushed anyway, the file never had the column, so the predicate cannot see the filled default. Spark would filter the wrong thing.

This PR stops pushing file-level predicates for omitted defaulted fields. Spark evaluates those filters on the filled values after read. Empty projections no longer crash.

Also: TypeUtil schema copies (AssignFreshIds / PruneColumns / ReassignIds) now keep initialDefault via NestedField.from(field), so table-create and SELECT round-trips do not drop defaults.

Stacked on #268 → #267.

Testing Done

export JAVA_HOME=$(/usr/libexec/java_home -v 11)
./gradlew -DsparkVersions=3.1 -DscalaVersion=2.12 -DhiveVersions= -DflinkVersions= \
  :iceberg-api:test --tests 'org.apache.iceberg.types.TestTypeUtil' \
  :iceberg-orc:test --tests 'org.apache.iceberg.orc.TestExpressionToSearchArgument' \
  :iceberg-spark:iceberg-spark-3.1_2.12:test \
  --tests 'org.apache.iceberg.spark.source.TestSparkOrcInitialDefaultScan' \
  --tests 'org.apache.iceberg.spark.data.TestSparkOrcReaderForFieldsWithDefaultValue' \
  -x generateGitProperties

• Filter on only the omitted defaulted column — no crash, Spark sees the filled value
• Mixed physical + defaulted predicates
• Defaulted scans stay on the row reader even with vectorization enabled
• initialDefault survives TypeUtil copies