2. Putting references to the TLA+ spec in the code could be nice and useful.
3. You should realize that the benefit from a powerful tool like TLA+ comes when it helps you reason about things that are hard for you to do, and those are hardly ever at the code level. So while using a code-level tool like Why3 to prove some routine is cool, it is hardly ever done in practice, because that's hardly every where the problems you need help with are. Those are usually found in complex interactions between machines, or your program and the user etc.. So often TLA+ is most useful when the specification is so far above the code level that a direct link is not entirely useful.
Having said that, you could use TLA+ to actually prove at the code level by compiling your code to TLA+, as has been done with Isabelle or Coq. This has been done for TLA+ with C and Java, but in practice, code-level verification is not where you'll get the most bang for your buck.
E.G. Do you have a "proof" project and a the code in different directories with comments in the TLA+ pointing to the files in the actual code?
It would be incredibly nice if I could put TLA+ proofs in the comments next to every function or module I am working with and then choosing which of them to be proved. But that seems to be way beyond what the toolset allows you to do.