a subset (or subsets) from a superset. If you don't have a huge amount you can do that.
Else you'll need a seeded approach.
There might be more solutions or none. Can't calculate them as it is kind of an non deterministic problem.
Tricky issue and asked a lot.
Well, I looked at what seems to be a similar issue just recently. It is a complex task, and is more complicated depending on the requirements of the final polygon (does it need to be as compact as possible, for example?)
Basically FME has no single transformer intended to carry out this sort of function.
However... the task could be described as a form of "bin packing" or "clustering" (I don't think it's really clustering, but a lot of information on the subject uses that) and if you Google those terms you might find some algorithms that you can replicate in FME, or Python-based scripts that will run in FME.
For example, this paper has some useful information.
What you could try is creating a minimum spanning tree (see this blog post) and from there group together features with a tree connection until the maximum value is reached.
This is something that our developers will looks into shortly, but I don't know what the time frame will be exactly. Almost certainly it won't be FME 2019.