v9.2.0

🧪 Experimental pyatlan_v9

• IDE autocompletion: Replaced custom AST-based lazy loading with lazy_loader + Pkl-generated .pyi stub file. All 860+ asset classes now have full IDE type hints and autocompletion while keeping lazy import performance.
• Pkl-generated __init__.py and __init__.pyi: The Pkl renderer now aggregates all typedefs (via import* glob) to generate __init__.py (with lazy_loader.attach() mapping) and __init__.pyi (with explicit re-exports for IDEs) directly — no post-sync Python script needed. Hand-written types from _init_manual.py are included by reading the file at generation time.
• Model updates: Regenerated all v9 models — GTC anchor regularization, entity lineage fields (depth, immediate_upstream, immediate_downstream), QuickSight useLocalTypeAsPrefix, fully Pkl-generated DataQualityRule, and referenceable field descriptors via overlays.
• type_name serialization fix: type_name field default changed to UNSET so omit_defaults=True never omits typeName from API requests.
• Simplified transform.py: Removed _normalize_camel_key() and related camelCase abbreviation handling — msgspec structs use explicit field(name=...) mappings. Asset registration now uses cls.__name__ directly.

🥗 QOL Improvements

• Regenerated latest typedef models.
• Improved Claude Code review skill and added auto-review trigger on PR open.

⛑️ Breaking Changes

• Policy duplicate detection during retries: When a policy creation request times out but actually succeeds on the backend, the SDK now detects the existing policy instead of blindly retrying and creating a duplicate. On retry, the transport layer searches for an existing AuthPolicy with the same name and persona GUID — if found, it returns a mock response containing the existing policy. Code that relied on retries always creating new policies will now receive the existing policy instead.

Full Changelog: 9.1.0...9.2.0