Set executor port from model deployment file

[chengcheng-pei]

now, the executor port is determined by ENV variable in seldon-core-operator and default port:

seldon-core/operator/controllers/seldondeployment_engine.go

Line 123 in 4446bac

func getExecutorHttpPort() (engine_http_port int, err error) {

But, we need the flexibility to set the executor port from model deployment resource file.
e.g. when deploying xgboost, we set port to 5001. when deploying tf, we set port to 8000.

Otherwise, we need update service too. for executor, the service exposes 8000 for rest and 5001 for grpc, which is good enough.

[ukclivecox]

The executor runs on a single port. This is separate from any model port for http or grpc which can be defined in the graph section of the spec. Not sure this is required. Please reopen if use case is more specific.

[chengcheng-pei]

now, the only place to define the executor port is in seldon-core-operator env.

it is not very flexible to use. we still try to use 5001 as grpc port, and 8000 for rest.
but after seldon-core installed, we cannot use two ports for grpc and rest.