For instance, when debugging a simple Python utility, I will often put the steps it goes thru into individual functions and tie them together with a main() function normally called via "if __name__ == '__main__':main()".
That allows me to import the program as a module into a REPL session, from which I can call the functions individually and examine the resulting outputs. If something's not right, I can alt-tab to the editor, make a change, save the program, and use the reload() function in the REPL. Sometimes, the REPL-obtained data is all I'm looking for, and I never get around to running the code from the command line.
When testing with a debugger I setup my program to run the same tests, read the same input etc. In this way I can use the debugger, find and issue, change the code, set a break point, re-run the program and I'm right back where I left off. This sounds functionally similar to using a REPL.
The IDE part is not true at all, why do people keep saying this? LiteIDE (aka golangide) works great, I use it all day everyday. It has syntax highlighting, autocomplete, gofmting, building, debugging, etc and is very lightweight.
Downloads: https://code.google.com/p/golangide/downloads/list Source: https://github.com/visualfc/liteide
As for the merit if REPL over something like the go playground (http://play.golang.org/) or a debugger, I can't say. Most of the code I wrote before Go was C/C++/C#/Java.