Make DebugInfo edges per-call rather than per-callee

[topolarity]

This is useful for when you need to, e.g., trim the "common prefix" of two stacktraces that were both taken from inside an inlined function call. I'm hoping to use this property to provide a way to get only the "popped" frames inside a catch handler, but that requires us to understand when an inlined frame was popped, as in:

@inline function foo()
    try
        error("foo")
    catch err
        Base.showerror(stderr, err, Base.catch_backtrace())
    end
end
bar() = foo()

The frames would be:

Stacktrace:
 [1] error(s::String)
   @ Base ./error.jl:44
 [2] foo
   @ ./REPL[1]:3 [inlined]
 [3] bar()        # <- this hasn't been "popped" yet, we want to exclude it from the debuginfo expansion
   @ Main ./REPL[2]:1

To know to trim off bar() (since it hasn't been popped yet), I want to record the ip at catch entry and then have our stacktrace expansion logic remove any frames common with the catch entry.

In order to trim that common prefix correctly, we need to distinguish this case from:

try foo() catch foo() end

which has the same line number and callee, but is a different call.

[topolarity]

On second thought, I don't think this change is needed

The catch ip should yield a strict subset of inlined frames that we got from the try (where we threw), so it should be enough to record the ip on catch entry and use that to count how many inlined frames should be left "unpopped"