Do not propagate taint through a channel that is tainted in one select branch and read from in another branch

[mlevesquedion]

Consider the following test case:

func TestTaintedAndSinkedInDifferentBranches(objects chan interface{}) {
	select {
	case objects <- core.Source{}:
	case s := <-objects:
		// TODO(211) want no report here, because objects is only tainted if the
		// other branch is taken, and only one branch can be taken
		core.Sink(s) // want "a source has reached a sink"
	}
}

Given the semantics of select, one of two things can happen:

1. We put a source on the objects channel
2. We get an interface{} value from the objects channel and we sink it

As indicated by the TODO, it is therefore not possible that the channel be tainted and that a tainted value is obtained from it in the same call to this function.

However, this code may still be unsafe, because the fact that one of the cases puts a source on the channel indicates that the channel may contain sources in general. Consider e.g. a case where this function is called concurrently in two different goroutines. Such a case would be unsafe code. Handling this behavior in full generality within the scope of a single function would require us to model every interaction with the channel. Alternatively, this could be handled via interprocedural analysis (#99).

This issue is somewhat related to #161 and #210, in the sense that it is related to a behavior where we may want to leverage