Support collection type of MetadataField in Vector Store

In some vector stores like Azure AI Search, the metadata fields could be collection type, and support filter collection metadata by users.

Say keywords field, in Azure AI Search, we can define this field as a collection type of string

new MetadataField("keywords", SearchFieldDataType.collection(SearchFieldDataType.STRING))

And then query the vector store by setting the filter that any documents that have at least one of the keywords 'cat' and 'dog' using OData query:
keywords/any(t: search.in(t, 'cat, dog'))

Expected Behavior

Return all documents with keywords contains 'cat' or 'dog'.

Current Behavior

It's not supported, only single value metadata types and 'in' operator are supported in Azure AI Search.

Context

There are many scenarios we need set a group of keywords or tags into one metadata field, and search with filtering the keywords or tags.

No files or tests are named. Start by searching for MetadataField and the Azure AI Search filter implementation, then trace how metadata types and the in operator are translated. Done means collection-valued fields can be defined and filters such as keywords/any(t: search.in(t, 'cat, dog')) return matching documents.