The implementation I propose is grouping the different dictionaries sent to create
by their "schema" in order to use them in only one query (I intend to make the implementation better by taking into account the maximum length of a SQL command later on).
In order to preserve the order of creations, grouping is done "localy" ie if the vlist
is something like [{'a': …}, {'b': …}, {'a': …}]
three inserts will be generated anyway.