Revisit assumptions about property capture logic when re-evaluating a binding

For both bindable and non-bindable properties, we keep a list of guards / triggers, but (roughly) only reuse them if they appear in the same order.

We also capture properties multiple times if we are using them more than once in a binding, which seems wasteful (even if we only re-evaluate the binding once).

Maybe keeping a hash of object/propertyindex to guard/trigger would turn out to be more efficient. Or throwing the existing bindings / triggers away, and relying on the pool allocator to be fast.