[zhuanshenbsj1]

…BLOB fields on PK tables

Purpose

In Java Paimon (SchemaValidation.validateBlob), BLOB columns are supported on both append-only tables (primaryKeys.isEmpty(), requiring row-tracking. enabled = true and data-evolution.enabled = true) and primary-key tables (!primaryKeys.isEmpty(), with dedicated primary-key BLOB validation rules in validatePrimaryKeyBlob).

However, in pypaimon:
1. SchemaManager._validate_blob_fields unconditionally required row-tracking.enabled = true and data-evolution.enabled = true whenever a table had any
BLOB column, which prevented creating any primary-key BLOB table from Python (since row-tracking.enabled cannot be enabled on primary-key tables).
2. FileStoreWrite._create_data_writer blocked all primary-key tables containing any BLOB column, even when all BLOB columns were inline-stored (blob-descriptor-field or blob-view-field) and stored directly as BYTES inside normal Parquet/ORC/Avro data files without .blob sidecar files or rowIdtracking.

This PR aligns pypaimon schema validation with Java and enables creating primary-key BLOB tables, as well as writing and upserting inline BLOB fields
(blob-descriptor-field and blob-view-field) on primary-key tables in Python.

Tests

- `pytest paimon-python/pypaimon/tests/schema_manager_test.py` 
- `pytest paimon-python/pypaimon/tests/blob_table_test.py`

[JingsongLi]

The Python/Java parity and inline PK BLOB write/upsert use cases are useful. I found one schema-compatibility issue below. Verification: schema_manager_test.py and blob_table_test.py passed with PyArrow 19.0.1 (115 passed, 1 skipped, 36 subtests passed); additional format/bucket upsert and delete checks passed. Three Python-created partial-update BLOB configurations were independently rejected when loading the same persisted tables through Java FileStoreTableFactory.create.