feat(shared-shredding): Expose shared-shredding utilities

[dalingmeng]

Purpose

Linked issue: close #342

Expose shared-shredding schema metadata utilities for downstream integrations.

This change introduces MapSharedShreddingSchemaUtils as a public schema-level utility for shared-shredding MAP layout. It provides:

• Physical sub-column naming contract, including __field_mapping, __overflow, and __col_i.
• AttachMetadataToSchema to attach serialized shared-shredding metadata to Arrow fields.
• ExtractMetadataFromField to parse shared-shredding metadata from an Arrow field.

The internal MapSharedShreddingDefine continues to own metadata keys, while public callers can reuse the schema contract and metadata read/write helpers without depending on internal src/ headers.

Tests

• MapSharedShreddingSchemaUtilsTest.*

API and Format

This change affects public API in the include dir.

New public header:

• include/paimon/shredding/map_shared_shredding_schema_utils.h

New public API:

• MapSharedShreddingSchemaUtils::AttachMetadataToSchema
• MapSharedShreddingSchemaUtils::ExtractMetadataFromField

This change does not introduce a new storage format or protocol. It exposes existing shared-shredding schema/metadata conventions through public utilities.

Documentation

No.

Generative AI tooling

Generated-by: Qoder

[lxy-9602]

+1

[zjw1111]

Follow-ups on the newly exposed schema utils. See inline comments for the type-guard and test-gap items.

One doc fix on the PR description:

• The new public header path is include/paimon/data/shredding/map_shared_shredding_schema_utils.h, not include/paimon/shredding/....
• "New public API" omits LogicalToPhysicalSchema, which is also exposed publicly.

Please update the description to match.

[zjw1111]

LogicalToPhysicalSchema is now a public API, but the underlying impl does static_pointer_cast<arrow::MapType>(field->type()) without checking the field is actually a MAP. Internal callers only pass known MAP columns, but an external caller passing a non-MAP field name in field_to_num_columns would hit UB instead of a clean error. Please add a type guard (e.g. field->type()->id() == arrow::Type::MAP) and return Status::Invalid otherwise.

[zjw1111]

Test gap: all cases pass valid MAP columns. Please add a case for LogicalToPhysicalSchema with a non-MAP field name in field_to_num_columns (covers the type-guard above), and for ExtractMetadataFromField where the field exists but has no shredding metadata.

[zjw1111]

The description uses close #342, which auto-closes the issue on merge. This PR only exposes the utilities and shouldn't close #342 — please change it to a non-closing reference (e.g. Related to #342).