Engine: Do not call serializer for None values

[sphuber]

Fixes #5693

Any port that defines a serializer, will have that serializer invoked
when a value is passed to this port. The invokation was correctly
skipped if no serializer was defined or if the value already was a
Data node (in which case serialization is no longer needed), but it
was also called when this value is None.

It arguably doesn't make sense to serialize a value that has not been
defined. This behavior would cause problems for processes that define a
port with a serializer but that also accept None. In this case the
serializer would nevertheless be called and, in the case of the
to_aiida_type serializer for example, would except.

[sphuber]

I added two commits for this PR. Strictly speaking, only the first is necessary to fix the immediate bug, and I think that is the correct fix. But the second commit by itself would have also fixed the immediate cause for the bug in question, however, it would not be as general a fix. I still think that it makes sense to have to_aiida_type not choke on None.

[unkcpz]

@sphuber thanks!

Strictly speaking, only the first is necessary to fix the immediate bug, and I think that is the correct fix. But the second commit by itself would have also fixed the immediate cause for the bug in question, however, it would not be as general a fix

Do you mean either one of the commits can solve the issue?

I still think that it makes sense to have to_aiida_type not choke on None.

I am not sure I understand this, is this addressed by the first commit?

[sphuber]

Do you mean either one of the commits can solve the issue?

What I mean is that the current situation is: if a serializer is defined on an InputPort, that serializer is called, even if the value of the port specified is None. If the serializer does not know that it can receive None and does not deal with it, it might except.

I am not sure I understand this, is this addressed by the first commit?

This is the case for to_aiida_type, which raises when it gets a None value.

So there are two approaches to fix this:

1. Simply have to_aiida_type accept None and simply return None (this is the second commit)
2. We think that this is a broader problem and a serializer should simply never be called when the value is None (this is the first commit)

So the second commit just fixes the problem for one specific serializer, whereas the first commit addresses the problem for all potential serializers. I am just not sure if there are serializers out there now that somehow rely on None being passed. I doubt it to be honest, but we would be changing that behavior.

[unkcpz]

In my opinion, I think serializing None makes not much sense. The None set in the argument is more like a placeholder than the data to be serialized. If 1st commit already solve the issue I think it is enough and it should not go further and try to serialize it.

[sphuber]

Thanks @unkcpz . Unless there is anyone else that has a strong opinion, I will then remove the second commit and merge this by Friday.

[sphuber]

@unkcpz I have removed the second commit. Could you review and approve please?

[unkcpz]

Just have one minor request for the docstring. I assume it is my language issue, so I approve it.