Allow multi-model repositories for Tensorflow Serving

[ukclivecox]

Check if we can relax the constraint that the model path needs to exist when using Tensorflow protocol

seldon-core/executor/api/rest/server.go

Lines 315 to 320 in aad5615

	if r.Protocol == api.ProtocolTensorflow {
	if modelName != "" {
	if graphNode = v1.GetPredictiveUnit(&r.predictor.Graph, modelName); graphNode == nil {
	r.respondWithError(w, nil, fmt.Errorf("Failed to find model %s", modelName))
	return
	}