Fix trust region exit criterion

In trust_region.m we have a problem with trying to encode three pieces of information in a binary variable check

The problem is error code -3 (Trust region became uselessly small). c01352ea tried to prevent the solver from accepting error code -3 as a solved steady state. But this breaks the loop in dynare_solve.m:

    for i=length(r)-1:-1:1
        if options.debug
            disp(['DYNARE_SOLVE (solve_algo=2|4): solving block ' num2str(i) ', of size ' num2str(r(i+1)-r(i)) ]);
        end
        [x,info]=solver(func,x,j1(r(i):r(i+1)-1),j2(r(i):r(i+1)-1),jacobian_flag, ...
                        options.gstep, ...
                        tolf,options.solve_tolx, ...
                        options.steady.maxit,options.debug,varargin{:});
        if info
            return
        end
    end

which relies on being able to continue with the loop when the code internally was -3 (as ~(-3)=0), i.e. we discard the information that -3 is a problem. But further below we have the final check

    if max(abs(fvec)) > tolf
        [x,info]=solver(func,x,1:nn,1:nn,jacobian_flag, ...
                        options.gstep, tolf,options.solve_tolx, ...
                        options.steady.maxit,options.debug,varargin{:});
    end

Here, we would need to know whether a 0 in info comes from info = (1) (everything fine) vs. info=(-3) (still a problem after the last iteration).

My proposed solution would be to augment both trust_region.m and solve1.m with a third output argument exitflag that encodes the full integer error code. That way we would only need to check for this argument in the final check of dynare_solve.m