fix(graph_dbs): sanitize nested metadata before Neo4j writes

[kyan-du]

Summary

• sanitize nested metadata values before writing Neo4j node properties
• apply the same sanitization in both Neo4j enterprise and community graph DB implementations
• add a regression test covering skill-memory style nested metadata

Problem

Neo4j node properties only accept primitive values or arrays of primitives. Some MemOS flows (especially SkillMemory and feedback-related flows) can produce nested metadata like:

• scripts: dict
• others: dict
• nested objects under info
• arrays containing dict items

Those values can reach SET n += $metadata and trigger errors like:

• Property values can only be of primitive types or arrays thereof
• CypherTypeError: Map{} encountered

Fix

This PR adds a small sanitization layer before Neo4j writes:

• keep primitive values as-is
• keep arrays of primitives as-is
• serialize dict values to JSON strings
• serialize nested list items that are dict/list values

This keeps metadata readable while making it safe for Neo4j property storage.

Validation

• reproduced the issue locally with SkillMemory-style metadata
• verified the sanitized metadata no longer contains Neo4j-invalid map values
• verified a real /product/add flow successfully created SkillMemory without the previous Neo4j Map{} error

[kyan-du]

This PR addresses the root cause described in #1172 and partially in #1122.

The existing _flatten_info_fields() fix (from #1122) handles the top-level info dict, but other dict-typed fields (scripts, others, and nested objects inside info) still reach Neo4j as Map{} values.

This PR adds a universal sanitization step that catches all remaining cases.