refactor(memory): enhance extraction ontology and add assistant pruning graph support
- Expand entity type ontology with detailed definitions, examples, and notes
(merged types: 地点设施, 物品设备, 产品服务, 软件平台, 角色职业, 知识能力, 偏好习惯目标, 称呼别名, 智能体)
- Add relation ontology taxonomy with 15 predicate categories and usage rules
- Strengthen reference resolution rules: resolve pronouns before extraction,
skip unresolvable references entirely
- Add guidelines to avoid extracting abstract propositions, emotions, and
low-value entities (effort/reward/success patterns)
- Add 7 new extraction examples covering edge cases
- Add AssistantOriginal/AssistantPruned node models and graph persistence
(PRUNED_TO and BELONGS_TO_DIALOG edges, Neo4j indexes and constraints)
- Add graph_build_step.py for building graph nodes/edges from DialogData
- Update write_pipeline.py to pass assistant pruning nodes/edges to graph saver
- Update data_pruning.py with related preprocessing changes
7747ed7ac1