fix: normalize business knowledge vector IDs

[March-77]

What changed

• serialize new businessTermId vector metadata as a string
• query both the normalized string representation and legacy numeric metadata
• delete both metadata representations when business knowledge is updated or removed
• add a regression test backed by the actual SimpleVectorStore filter implementation

Root cause

SimpleVectorStore evaluates metadata filters through SpEL. Numeric values in an IN expression are parsed as Integer, while business knowledge IDs were stored as Long, so equality always failed and no business terms were recalled.

Validation

• ./mvnw -pl data-agent-management -Dtest=DynamicFilterServiceTest,DocumentConverterUtilTest,BusinessKnowledgeServiceImplTest test (58 tests)
• ./mvnw -pl data-agent-management spotless:check checkstyle:check
• git diff --check

Fixes #460

[March-77]

已合并最新 main 并解决冲突。新写入的业务术语使用字符串 ID，删除和过滤仍兼容旧的数值 ID。相关测试 72 个通过，格式和 Checkstyle 也通过。