Slow tests are tests that aren’t run. Having unit tests that depend on databases and APIs are slower, increase the deployment cycle and usually end up being skipped.
Trying to make dependencies substitutable is overengineering and leads to poor design and gives you more work when implementing unit tests.
How is it a unit test if it can fail due to other dependencies?
Hiding the source path of a dependency is a terrible anti-pattern. When it comes to programming, there are few things more horrible than not being able to determine where some buggy piece of logic is located. I cannot imagine any use case where that would be a fair tradeoff.
If you are only using dependency injection to enable unit testing, finding the real implementation is trivial - it’s the only concrete implementation of interface to the constructor.
Any kind of injection of dependencies should be done via an explicit method/constructor parameter. Sometimes it means that the dependency instance has to traverse a few classes in the hierarchy, but that's way better because at least you can unambiguously track where the dependency came from.
Again if you have only one concrete implementation and you’re using a DI framework, finding the implementation of an interface is trivial with any halfway decent IDE.