Declare the embedding model as metadata on VECTOR columns

A VECTOR(N) column stores embeddings, but their semantics depend entirely on the embedding model that produced them.
Vectors from different models are not comparable — a similarity search over a column containing mixed-model embeddings silently
returns plausible-looking but meaningless results.

Today the model must be tracked out-of-band: naming conventions, companion tables, or application code.
In practice, multiple independent writers touch the same column.

The MariaDB MCP server, ETL jobs. Notably, the MariaDB MCP server already accepts a model_name parameter in create_vector_store,
but this information is not persisted as queryable schema metadata, so no other client can see it.
Only the schema is a shared source of truth.

Allow recording the embedding model name and version as metadata of a VECTOR column.

This gives tools and applications a way to check compatibility before inserting or searching,
and makes model migrations (re-embedding a table) auditable: the attribute documents the current state, and tooling can find stale columns.