v3.0.1

✨ New Features

• Added DataType.TEXT handling across schema validation, result decoding, and BulkWriter paths.
• Added get_replicate_info() to MilvusClient and AsyncMilvusClient.
• Added sync and async dump_messages() support for data-salvage workflows.
• Added role and user descriptions, including alter_role() and role-description output from describe_role().
• Added search aggregation and correctly forward group_by_fields.
• Added FunctionChain APIs, forward every specified stage during search, and added the XGBoost expression helper.
• Added alter_collection_schema() and support for dynamically adding struct fields.
• Support for external field mappings and add-field fallback for external collections.
• Added drop_function_field(), support for MinHash function fields, and require bound index parameters when adding a function field.
• Consolidated logical type metadata across search, insert, array, row, and vector paths to remove inconsistent type-specific behavior.
• Accept float inputs for FP16 and BF16 vectors.
• Improved connection recovery with atomic sync and async reconnect swap channels, protection for active RPCs, and retired-channel reclamation.
• Preserved configured connection timeouts during recovery.
• Mapped only connection-setup RPC failures to CONNECT_FAILED and exposed error classification through MilvusException.
• Preserved nullable vectors in CSV BulkWriter output, supported struct-vector fields, and skipped payloads for null struct rows and null struct-vector subfields.
• Resolved binary-vector search inside struct arrays, preserved struct-array fields reconstructed from schema dictionaries, and supported partial updates for struct-array fields.
• Materialized array query results as standard Python lists.
• Improved Volume uploads with safer per-upload options, lower default concurrency, byte-level progress, and request-local concurrent-upload state.
• Preserved remote upload prefixes on Windows.
• Rejected oversized or non-positive compaction target_size values.
• Rejected booleans as sparse-vector values or indices and improved validation errors.
• Rejected empty binary-vector lists with a parameter error instead of leaking an IndexError.
• Allowed null values in an auto_id primary-key DataFrame column during insert.
• Correctly unpacked async schema results before collection optimization.
• Migrated development and CI workflows to uv, used Hatchling-backed version resolution, and retained the intended release-branch RC version semantics.
• Isolated protobuf-generation dependencies and pinned the final Milvus 3.0 proto, regenerating the Python bindings.
• Classified maintained test suites and updated async, struct-field, and BulkWriter examples.