One method involves incorporating dependency injection, while the other employs the service locator design pattern. Dependency injection comes with a variety of benefits:
It clearly defines a component's dependencies. For instance, examining a controller's constructor function reveals precisely what dependencies it relies on. In contrast, when using a service locator, one may not always know, as the controller could call upon the service locator at any point.
Unit testing is simplified. Utilizing the $controller service allows for instantiating a controller and providing it with mock objects via the locals parameter. This proves more convenient compared to creating numerous AngularJS services or factories for resolution by $injector. The situation worsens when combined with point #1: there may be unidentified dependencies essential for supplying all necessary mocks.
However, service locator does present some flexibility. For example, code can check if a service exists before utilizing it, like so:
if ($injector.has('serviceA')) $injector.get('serviceA').doSomething()
else doSomethingElse()
In contrast, with dependency injection, encountering an error will occur if serviceA isn't available during your component's construction.
Regarding refactor ease, adding or removing parameters from a constructor function shouldn't pose significant difficulty. Additional tools such as ngAnnotate streamline declaration DRYing, as a comment on your query mentions.
Ultimately, sticking with dependency injection seems prudent, saving the use of service locator for situations where it's truly indispensable.