Fix issues with AbstractCodeWriter state stacks

[mtdowling]

Code writer context inheritance wasn't working in the previous implementation of the stack based inheritance system. We were able to resolve context from the current state and parent state (because of eagerly copying states and flattening them per/state), but we weren't able to look at parent states that used a CodeSection. Methods of a CodeSection state can be accessed in named template labels. However, when another state is pushed after setting a CodeSection value for a state, code writer wasn't crawling the state stack to see if previous CodeSections could provide a value for a named label that wasn't explicitly set as parameter in the context map. This was because instead of looking up values lazily by iterating over the state stack, we were making eager copies of different pieces of states. This matters when using things like for loops in templates that push an implicit state to capture the current loop iteration state.

This commit changes our approach to now actually use the stack itself and iterate over states when attempting to resolve context keys, formatters, iterators, and CodeSection values.

Some tests were added to ensure the behavior of the writer is the same. For example, when using inheritance based lookups, removing a context key now means we need to set a context key to an explicit null value to emulate the previous behavior. If we didn't do this, and we just called Map#remove(), the context key may have been set by a parent state, making the remove call do nothing (whereas the previous implementation would actually take effect).

Background

• What do these changes do?
• Why are they important?

Testing

• How did you test these changes?

Links

• Links to additional context, if necessary
• Issue #, if applicable (see here for a list of keywords to use for linking issues)

---

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.