Support blob v2 source columns in ADD COLUMNS FROM

[hfutatzhanghb]

Context

Follow-up from #764 (comment).

Current behavior

ALTER TABLE ... ADD COLUMNS ... FROM ... does not accept a blob v2 source column when the target configures <column>.lance.encoding = blob.

Spark reads a blob v2 column as a descriptor struct. AddColumnsBackfillExec passes the query schema through SchemaConverter.processSchemaWithProperties before AppendData is built. Because the source field is a struct rather than Spark BINARY, addBlobMetadata rejects it with:

Blob column '<name>' must have BINARY type

The rejection happens before LanceBlobV2CopyThroughRule can rewrite descriptor reads into blob copy references. By contrast, direct INSERT INTO ... SELECT between blob v2 tables supports descriptor-to-blob copies and is covered by BaseBlobV2CopyTest.

Question

Should ADD COLUMNS ... FROM also support direct blob v2 source columns, analogous to INSERT INTO ... SELECT?

Possible direction

• Preserve or expose blob v2 field metadata during AddColumnsBackfillExec schema processing.
• Let the command reach an AppendData plan that LanceBlobV2CopyThroughRule can rewrite, or add an equivalent copy-through path for this command.
• Keep the requirement that _rowaddr and _fragid address target rows.
• Add integration tests for copying blob bytes from a blob v2 source table into a newly added blob v2 target column.

Current workaround

Copy blob columns between existing tables using direct INSERT INTO ... SELECT, as documented in docs/src/operations/dml/insert-into.md.