Consider introducing `citus.force_max_executor_pool_size`

We currently have `citus.force_max_query_parallelization`, which forces one connection per placement. That might not be ideal, as it relies on shard count, and ignores `citus.max_adaptive_executor_pool_size`. Instead, if we had `citus.force_max_executor_pool_size` that would be much handy for users.

Note that we mostly introduced `citus.force_max_query_parallelization` for two things. First, it is very useful in the tests, we force executor to open multiple connection to test certain features. Second, help users in the mentioned cases here:http://docs.citusdata.com/en/latest/develop/api_guc.html#citus-force-max-query-parallelization-boolean

I presume `citus.force_max_executor_pool_size` would work better for both.