Fix: Prevent loss of domain information when template field is present in layout

[marthacryan]

Alternative fix to plotly/plotly.py#4794. Other fix is #7282.

Context:

• @alexcjohnson noted that Fix: Switch template edit type to plot #7282 didn't seem like the right fix because changing editType from 'calc' to 'plot' is more of a downgrade than something that should fix this issue.
• I pointed out this particular line should be resetting the _inputDomain, but that this line was returning in diffFlags before it reached the yaxis to reset it
• Alex said that the fact that we're resetting _inputDomain in the changed function was actually incorrect. It should be reset in its own separate crawl of the layout.
• We decided to put the reset of _inputDomain into the diffLayout function because it has access to both the old layout and the new layout.

[alexcjohnson]

I think nestedProperty(...).set(null) is actually equivalent to delete newFullLayout[key]._inputDomain

[archmoj]

@alexcjohnson Using delete may be slower comparing to setting to null. No?

[alexcjohnson]

I suppose that’s faster and should be equivalent. This is not a high volume operation so I wouldn’t be too concerned about it, but we can leave it if you prefer.

[marthacryan]

I was using this syntax for consistency with the other places in the codebase that delete this field, but I don't have a strong opinion on which we use.

[emilykl]

@marthacryan Maybe worth leaving a comment explaining why we do nothing in the else case? (I know we discussed but... for posterity)

[marthacryan]

Ah yeah good point Emily!

[marthacryan]

@alexcjohnson I added a comment explaining the else case based off of this explanation from you about that case:
"this situation is perhaps actually ambiguous: did you pass back in the previous layout as modified by Plotly.js, so we should assume you really wanted _inputDomain to still apply? or did you explicitly set the new domain and it just happened to match the old final calculated domain - in which case we should clear _inputDomain. I guess generally oldDomain will be some random floating point number so it's unlikely you matched it exactly. better to assume you passed back in the same layout object and change nothing in this case."

Let me know if you have any corrections!

[archmoj]

@marthacryan Thanks very much for another excellent PR. 🏆
Please use Fix: prefix in the PR title and also add a draft log.
You may also consider revisiting the PR title so that it reflects the fix from the user's perspective.

[archmoj]

@marthacryan Could you please add a test to lock the bug you fixed in this PR?
You may add a it block (and then if you like to just run the new test, temporarily change it from it to fit) in test/jasmine/tests/plot_api_test.js and use npm run test-jasmine plot_api.
Obviously this new test should fail before your changes and should pass after the changes.
Thank you!

[marthacryan]

@marthacryan Could you please add a test to lock the bug you fixed in this PR? You may add a it block (and then if you like to just run the new test, temporarily change it from it to fit) in test/jasmine/tests/plot_api_test.js and use npm run test-jasmine plot_api. Obviously this new test should fail before your changes and should pass after the changes. Thank you!

@archmoj Done! I added it to the axes_tests.js file instead because there were a lot of relevant tests and utils in that file. It fails on master but passes on this branch.