And insert its definition if needed in source-to-source mode.

Due to the way Clang tracks locations for the `_Pragma` operator, its
end location cannot be rewritten.  An executable directive (the new
`acc update` is the first) has no associated statement to use for an
end location, so an executable directive in a `_Pragma` operator
directive cannot be rewritten.  Without this patch, the diagnostic
claims the directive appears in a macro in this case.  This patch uses
a better diagnostic for that case.

This patch also extends `-fopenacc-print` testing for executable
directives and updates the status doc.

In OpenMP, these are called stand-alone directives.  That is, they
don't have associated statements.  For example, `acc update`.

Without this patch, Clacc source-to-source mode when requested by
`-fopenacc-print` (but not `-fopenacc-ast-print`) often fails asserts
or mangles text when certain parts of OpenACC constructs expand from
preprocessor macros or when the `_Pragma` directive form is used.
With this patch, Clacc reports error diagnostics for cases not yet
handled, and other cases are fixed.

Such an error diagnostic is reported when either the first token is
expanded from a macro (thus the `_Pragma` form is used) or the
associated statement must be rewritten but its last token is expanded
from a macro.

This patch also redesigns `ACCExecutableDirective::getConstructRange`
to support the above changes as well as upcoming changes that will
provide source location information via the OpenACC Profiling
Interface.

The second OpenMP directive wasn't indented properly in `omp` or
`omp-acc` mode unless the associated statements were printed
separately.

That is, when associated statements print separately, print comments
clearly delineating and associating the OpenACC and OpenMP
alternatives.

For example:

  int i;
  #pragma acc parallel
  #pragma acc loop vector
  for (i = 0; i < 4; ++i)
    ;

prints with -fopenacc-print=acc-omp as:

  int i;
  #pragma acc parallel
  // #pragma omp target teams firstprivate(i)
  // v----------ACC----------v
  #pragma acc loop vector
  for (i = 0; i < 4; ++i)
    ;
  // ---------ACC->OMP--------
  // {
  //     int i;
  //     #pragma omp parallel for simd num_threads(1)
  //         for (i = 0; i < 4; ++i)
  //             ;
  // }
  // ^----------OMP----------^

When associated statements print separately, the "discarded in OpenMP
translation" comment only prints for this directive and not nested
directives.  That inconsistency makes it confusing what happened for
nested directives that were discarded.

Specifically, as for -fopenacc[-ast]-print=omp-acc, if associated
statements print separately, print OpenACC associated statement in acc
mode not acc-omp mode so that you don't confusingly sometimes print
some OpenMP twice.

And add -fopenacc-ast-print with the old functionality.