Feature/attribute identifier respect key

[diba1013]

Before submission, please check that ...

• the code follows the Clean Code guidelines.
• the necessary tests are either created or updated.
• all status checks (Travis CI, SonarCloud, etc.) pass.
• your commit history is cleaned up.
• you updated the changelog.
• you updated necessary documentation within retest/docs.

Description

TL;DR This fixes that the GlobalChangeSetApplier accepts similar attribute differences.

For the return of one-click-maintenance (more details in #726), the GlobalChangeSetApplier
uses the AttributeDifference#identifier() to find equal differences, since AttributeDifference#equals per definition cannot be used since it would check the warnings too.

However, the identifier currently only takes the expected and actual values into account, ignoring the key attribute. That means, following differences are considered equal, although differing in keys:

border-bottom-color: expected=rgba(0, 0, 0, 0.26), actual=rgba(0, 0, 0, 0.32)
border-top-color: expected=rgba(0, 0, 0, 0.26), actual=rgba(0, 0, 0, 0.32)

This causes review to accept these attributes with a single click (which, currently, is not the expected behavior, as that could group completely different differences).

Technically, the above example is the same difference, because in the report the differences could be grouped to border-color: expected=rgba(0, 0, 0, 0.26), actual=rgba(0, 0, 0, 0.32), but since we cannot implement grouping (technically), this is unwanted behavior.

State of this PR

I do not know what the intentions were to exclude the key from the identifier (since I could not search through the history of the different repositories), but a quick usage search indicates that this should not have any breaking side-effects.