handle variable evaluation in one location

[jbardin]

Module variable evaluation was spread out over 3 locations, which led to
a skew in what value could be seen within a module, and what value could
be seen within a variable's own validation scope.

First, the graph transformer would create a fake input expression on the
fly for variables with no input, and load it with the default value.
This caused trouble during actual evaluation where the input expression
would be mapped to the wrong source location, and made it difficult to
determine if an input value was really assigned or not.

The variables nodes would then evaluate the expression, but not be able
to handle defaults, because there always appeared to be an input
expression set for the variable. Because defaults could not be handled,
the wrong value could be passed into the validation expressions for the
variable.

GetInputVariable from the evaluator would check again for defaults, but
was usually fooled by the fake expression value. Even when it could
assign a default, as in the case of nullable=false with a null input,
it was already too late for the variable validation block to handle it.

We can unify all evaluation handling into the nodeModuleVaraible exec
node. This way the context is always loaded with the correct final value
of the variable, so GetInputVariable is only concerned with fetching
that value to evaluate references to it, and the graph transformer is
taken out of the picture entirely.

Due to the possibility of uncovering other bugs while refactoring variable evaluation, this won't be backported to v1.1, however a more localized fix may be possible to mitigate the incorrect validation values.

Fixes #30307

[alisdair]

At first glance this looks like it overlaps with #29959, which is on my shame list for re-reviewing. Does that seem right to you?

[jbardin]

Oh, I thought that was only addressing the root variable problems 😞
At least Martin an I came to mostly the same conclusions about the sprawl of variable handling. I'll close this for now and join the other conversation in #29959.

[github-actions]

I'm going to lock this pull request because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active contributions.
If you have found a problem that seems related to this change, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.