Use sets for widening thresholds instead of lists

[sim642]

When widening thresholds are collected from the program, we do use a Set to avoid duplicates, but then it's just converted to a list with elements:

analyzer/src/cdomain/value/util/wideningThresholds.ml

Lines 103 to 108 in bb6f9aa

	let widening_thresholds = ResettableLazy.from_fun (fun () ->
	let set = ref Thresholds.empty in
	let set_incl_mul2 = ref Thresholds.empty in
	let thisVisitor = new extractConstantsVisitor(set,set_incl_mul2) in
	visitCilFileSameGlobals thisVisitor (!Cilfacade.current_file);
	Thresholds.elements !set, Thresholds.elements !set_incl_mul2)

Later, IntDomain uses this list by linearly searching for the next threshold x for u:

analyzer/src/cdomain/value/cdomains/intDomain.ml

Lines 563 to 568 in bb6f9aa

	let upper_threshold u max_ik =
	let ts = if get_interval_threshold_widening_constants () = "comparisons" then WideningThresholds.upper_thresholds () else ResettableLazy.force widening_thresholds in
	let u = Ints_t.to_bigint u in
	let max_ik' = Ints_t.to_bigint max_ik in
	let t = List.find_opt (fun x -> Z.compare u x <= 0 && Z.compare x max_ik' <= 0) ts in
	BatOption.map_default Ints_t.of_bigint max_ik t

This linear search is $$O(n)$$ while the same lookup in binary search trees would be $$O(\log n)$$. And actually Set.Make.find_first_opt from the OCaml standard library can be used exactly for this purpose (because that's what Sets are internally).
So we could avoid the intermediate lists and do the more efficient search on sets directly. Although I doubt it will make a notable difference to Goblint performance as a whole.