remove deprecated model_type param (embedding)

[mattf]

NVIDIAEmbeddings' constructor accepts a model_type parameter with value of either "query" or "passage"

using it changes the behavior of the standard embed_query and embed_documents methods.

for example, setting model_type="passage" and calling embed_query results in using the same behavior as not setting model_type and calling embed_passage.

this parameter is confusing to users and has been removed.

users should use embed_query when embedding queries and embed_passage when embedding passages.

if you would like to batch embed queries, please raise a feature request.

note: the query / passage distinction is only important for asymmetric embedding models

[dglogo]

lgtm! thanks for doing this