OptaPlanner: Chained immovable planning entities shouldn't move (regression from 6)

I just discovered this regression when looking at the Task Assigning example, but it seems to affect VRP, TSP and all other "chained variable" use cases too.

Impact: any use case of Vehicle routing where the user pins a visit, or any such use case with continuous planning, is now broken. The "pinning" or "history" aspect is partially ignored and those visits that should never change, do get changed by OptaPlanner.

Caused by a refactor done around September 2017, this affects DM 7.0 GA and is a clear regression against BRMS 6. For details, see PLANNER-1009.
I'll work on a fix for master first, but I suspect this will need to be backported and put in the first update, hence this triaging jira.