fix(vector-stores): add agent_id and run_id to Elasticsearch/OpenSearch default mappings

[jessai2026]

Linked Issue

Closes #3744

Root Cause

create_index() in both ElasticsearchDB and OpenSearchDB only declared user_id as a keyword in the metadata mapping:

# Before — only user_id is explicit
"metadata": {"type": "object", "properties": {"user_id": {"type": "keyword"}}}

mem0 scopes memories by three canonical entity IDs: user_id, agent_id, and run_id (see memory/main.py:292–297). When a memory with agent_id or run_id is inserted first, both Elasticsearch and OpenSearch dynamically infer a type for those fields from the first value seen. If the inferred type differs from keyword (e.g. ES may infer object when an enum-style value is encountered), subsequent term filter queries on metadata.agent_id or metadata.run_id silently return empty results or raise a TypeError.

Fix

Explicitly declare agent_id and run_id as keyword in the mapping for both stores, matching the existing user_id declaration:

# After — all three scoping IDs are explicit keywords
"metadata": {
    "type": "object",
    "properties": {
        "user_id": {"type": "keyword"},
        "agent_id": {"type": "keyword"},
        "run_id": {"type": "keyword"},
    },
},

Files Changed

File	Change
mem0/vector_stores/elasticsearch.py	Add agent_id, run_id keyword fields to create_index mapping
mem0/vector_stores/opensearch.py	Same fix — identical bug in parallel implementation
tests/vector_stores/test_elasticsearch.py	Assert all three ID fields are present as keyword
tests/vector_stores/test_opensearch.py	Same assertion

Notes

• No migration needed for existing indices — ES/OS PUT mapping allows adding new explicit fields to an existing dynamic mapping without reindexing.
• The OpenSearch file had the same pattern as reported in Incorrect Elasticsearch mapping: cannot get memory results #3744 for Elasticsearch; both are fixed together to keep the two backends consistent.