yeah, looks as if the way we're doing this (using PGVector) doesn't have any real documentation or API exposed for creating filters, or even querying additional columns within the schema (we're using QueryEngineTool and ToolMetadata) so that quite different.
so, the "work around" would be potentially to search "by date" using a standard DB query, build a list of document_id's that match the date, and then pass those to VectorIndexRetriever as a doc_ids parameter?