[multistage][feature] allow configurable stage count limit to reject queries if they are too complex

We should have the option reject SQL that are too complex (for example generating 10 stages execution plan) since they will simultaneously acquire execution resources and thus hogging up the entire cluster's resource.

This should be a cluster level configurable key for users to tune

Start by tracing the multistage SQL query planning and execution path, then identify how cluster-level configurable keys are defined and read. Determine where the execution plan's stage count is available and how an over-limit query should be rejected; done means the limit is tunable at cluster level and complex queries are rejected before acquiring excessive execution resources.

Written by the indexing model from the issue text.