Incorrect handling of init expression on set of sets

[bbannier]

I would expect the following code to either work at runtime or be outright rejected:

# foo.zeek
event zeek_init()
	{
	local vs: set[set[string]] = { [ set() ] };
	assert |vs| == 1;
	}

$ zeek -b foo.zeek
error in foo.zeek, line 4: assertion failure: sizeof vs == 1
fatal error: errors occurred while initializing

I originally saw this with an initialization of the form

local vs: set[count, set[string, bool]] = { [ 1, set() ] };

In a Slack discussion there was an argument that this should not be supported at all and rejected since values in a set should be immutable (which a set is not). I'd argue that the usage here which does not modify the contained values at all could still be supported; we could still reject mutating operations on vs at runtime.

[evantypanski]

I'm going to close this out since it's more "user misunderstanding" than an actual bug, since sets of sets are fully supported. It's just really really awkward with how these lists get expanded, but I don't think changing the list expansion behavior will ever be a good idea.

Not to say I love how these cases get expanded, but it's technically "correct" and hard to change.

(See #4137 for those details)