np.int indexing inconsistency between SQL and RX #338

When using an np.int as node_id, it fails at multiple places when the graph is an SQLGraph. It works fine on IndexedRXGraph. This in undesired because:

1. It is an inconsistency between backends
2. It forces to user to make sure to always to int(nod_id) to cast np.int to int (which is especially a problem in napari-based applications where circulating integers are often np.ints

Example:

g.has_node(a)              # True
g.has_node(np.int64(a))    # False   <-- silently wrong
g.filter(node_ids=[np.int64(a)]).node_ids()   # []   <-- silently wrong

Reason:
sqlite3 binds np.int64 through the buffer protocol as an 8-byte BLOB, which never equals an INTEGER column.