w is now a real and follows the rules:

- [0..1]: defines a percentage of limit loop where the kappa is not
  used. For example -p w:0.5 -p limit:100 gives 50 loop without kappa
  computation and 50 with kappa computation.

- > 1: defines a static number of loop without kappa computation. For
  examples -p w:50 -p limit:-1 gives 50 loop without kappa and +oo with
  kappa computation.

The value of each variables in the solution/initial vector is
initialized according to the init-policy-random parameter of the
bernouilli's law:

- 1 use the init-policy algorithm to affect 0 or 1.
- 0 use the toss up algorithm (init-random parameter and the
  bernouilli's law) to affect 0 or 1.

Add into solvers and core API a BQP mode in objective function.

We use a shared_ptr to share a string_buffer between baryonyx::result,
baryonyx::problem and baryonyx::raw_problem and move the ownership from
variable names, constraint labels from std::string into
std::string_view.

To improve the use in command line interface, graphical user interface
or R, we replace, only for computation output (ex.: remaining
constraint message, new solution found, etc.), the output message
stream using private functions debug(ctx, ...), info(ctx, ...),
warning(ctx, ...)  interface with callback mechanism in the
baryonyx::context class. (Closes: #28)

The optimize_functor needs only constant references to the affected
variables and variable names vectors.