Only the H2O leader node should be queried by the end user at any time (reproducibility). While running eg. on Kubernetes,
it is easy for to misconfigure services and let the traffic flow to non-leader nodes. If an H2O cluster is used
in such a way, the results provided to the user might be irreproducible/incorrect and the user has no way to easily find out.
To solve this issue, this filter blocks all incoming requests on non-leader nodes when H2O is configured to have
non-leader node access disabled.
Some APIs intended for internal use (mainly used by other products such as Sparkling Water) might still remain active.
All user-facing APIs, mostly the ones used by Python/R/Flow clients are disabled.