Fix faulty executor config serialization logic

[dstandish]

The bind processor logic had the effect of applying serialization logic multiple times, which produced an incorrect serialization output.

Resolves #26101.
Related: #24117

[uranusjr]

Would it be more reliable to check for __type and __var instead? (We could have that logic in serialized_objects instead as a is_serialized function)

[dstandish]

are you maybe commenting on the wrong line? here we're dealing with non-serialized object.

[uranusjr]

I’m thinking line 170 should be

if not is_serialized(val_copy):

with the function implemented as

def is_serialized(v):
    if not isinstance(v, dict):
        return False
    return Encoding.TYPE in v and Encoding.VAR in v

This can be used to unify the logic on line 187/188.

[dstandish]

so the thing is, re 170, i see what you are suggesting. but the reason we are checking specifically for pod_override here, is because we're not trying to serialize any and all executor configs. in general, they are simply pickled. but because of idiosyncrasies of the design of the k8s objects, it's safer to essentially convert them to json before pickling , which in effect is what the serializer does. basically, we just want to do this for k8s executor configs. if we wanted to do for all configs, your suggestion makes complete sense.