Refactor lancedb_record_batch_reader_from_arrow to improve error handling

[polywj]

Description

This PR refactors lancedb_record_batch_reader_from_arrow to return a LanceDBError code and an optional error message, aligning it with other fallible APIs in the library. Previously, this function returned NULL on failure, providing no context on why the Arrow import failed (e.g., schema validation errors vs. array import errors). This closes #7.

Changes

• API Breaking Change: lancedb_record_batch_reader_from_arrow now returns LanceDBError instead of a pointer.
  • Added output parameter LanceDBRecordBatchReader** reader_out.
  • Added output parameter char** error_message.
• Implementation: Updated Rust FFI implementation in src/table.rs to catch and report errors during Schema and RecordBatch import from the Arrow C ABI using handle_error.
• Documentation: Updated include/lancedb.h and docs/index.rst to reflect the new signature and memory management rules (caller must free the error message if provided).
• Examples & Tests: Updated examples/full.cpp, examples/simple.cpp, and tests/test_common.cpp to handle the new return type and properly call the Arrow release function if the import fails.

Breaking Changes

The signature of lancedb_record_batch_reader_from_arrow in include/lancedb.h has changed.

Checklist

• API updated in include/lancedb.h
• Implementation updated in src/table.rs
• Examples updated (examples/simple.cpp, examples/full.cpp)
• Tests updated (tests/test_common.cpp)
• Documentation updated (docs/index.rst)