Certify collaborator step (in case of TLS) internally registers it, need to separate them out

[noopurintel]

Describe the bug
Right now, in case of TLS, fx collaborator certify step internally registers collaborators.
This leads to confusion while running a non-TLS experiment where we need to perform collaborator registration separately as it isn't mentioned anywhere.

To Reproduce
Steps to reproduce the behavior:

1. Refer to https://openfl.readthedocs.io/en/latest/tutorials/taskrunner.html
2. Git clone openfl and install all the pre-requisites.
3. Create workspace using fx workspace create --template torch_cnn_mnist --prefix my_workspace
4. Move inside my_workspace folder
5. Modify the plan/plan.yaml and assign use_tls: false (default value is true)
6. Initialise the plan.
7. Skip the certify steps at aggregator and collaborators level.
8. Create 2 collaborators - keep the names different from usual collaborator1/2 to easily identify the problem.
9. Run fx collaborator create -n a1 -d 1 and fx collaborator create -n a2 -d 2
10. At this point, plan/data.yaml gets updated with two entries a1,1 and a2,2
11. Start the collaborators and aggregator.
12. Collaborators a1 and a2 will start and keep waiting for the aggregator whereas aggregator will result in below error.
    ValueError: Expected sequence membership, but a2 is not in ['collaborator1', 'collaborator2']

Expected behavior

1. There should be a step to add/register collaborators before we create them irrespective of TLS/non TLS scenarios.
2. With above step in place, the experiment should run smoothly without any error.