even thought the setjmp implementation for try statement won't be used in C++,
using properly scoped variables for them is beneficial, because we'll be able to
establish a 1:1 relation between nimrod scopes and C scopes.
Once we have that, we'll be able to keep the original names of local variables
to greatly improve the debugging experience (i.e. watch expressions and
hover tooltips will work).
This patch greatly improves the "step over" operation available in debuggers.
In practice, there are often 4-8 lines of C code generated for each nimrod line
Each such line will be responsible to a single step in the debugger that is
a) not expected by the user
b) taking the user to an incorrect line in the nimrod code
To keep this working, all code generation should use the rope formatting
facilities when producing new lines (i.e. $n and $N).
New semantics for the format string are introduced:
$n means "soft new line" that could be joined/broken when lineDir is enabled.
$N means "hard new line" that will always appear as a new line.
As an alternative to this approach, I also tested producing code like this:
#line "code.nim" 154
foo = bar; \
foo(bar) \
This is better for readability of the final output, but unfortunately it didn't
produce the desired result across all compilers/debuggers.
