[core] Support projected binary index manifest scans

[leaves12138]

What changed

• Add a reusable BinaryIndexManifestEntry view for index manifest records.
• Add projected binary scan overloads to IndexFileHandler, backed by IndexManifestFile.scan, following the existing BinaryManifestEntry scan pattern.
• Read only kind, partition, bucket, index type, and the global-index fields needed by planning, while reusing one mutable entry view.
• Expose schema field-name constants instead of duplicating on-disk field names in projections.
• Add focused coverage for the public handler API, projected field values, ADD/DELETE records, absent global-index metadata, and iterator reuse/clearing.

Why

Memory-sensitive planning only needs a small subset of each index manifest entry. Existing scans materialize complete IndexManifestEntry and IndexFileMeta object graphs, including fields that these planning paths never consume. A projected binary scan avoids those objects and reuses one mutable entry view across the iterator.

Impact

Existing index manifest serialization and readers are unchanged. The new API is opt-in and keeps the existing on-disk schema intact.

Validation

mvn -pl paimon-core -am -DskipITs -Dfast \
  -Dtest=BinaryIndexManifestEntryTest,IndexManifestEntrySerializerTest,IndexFileHandlerTest \
  -Dsurefire.failIfNoSpecifiedTests=false test

Result: 13 tests passed. Spotless and Checkstyle passed as part of the Maven build.

[JingsongLi]

+1