Is it truly possible to have a solid understanding of the interface and its use, though, without having a reasonable insight into, at least, the likely implementation of something?
I'm sure this isn't what you meant, but how about something like movement? I've met doctors who know anatomy like the (ahem) back of their hand, but they couldn't perform coordinated physical tasks to save their life.
Yes, that's the whole point of abstraction, to reduce cognitive load by ignoring irrelevant implementation details.
This is why some people don't like languages with big runtimes/standard libraries like C++.
I don't see how that relates?
You can understand the source of your libraries just fine, but still be grateful that they are usually hidden behind interfaces that let the compiler help you spot silly mistakes.
(OK, less so in C++, more so in eg Haskell.)
I trust that the interface, be it the C++ standard, or boost::whatever, is well enough tested that I don't need to know "how it works".
It would be nice to know how everything works but there is only finite time in the working day.