Relax type requirements for stopping criterion parameters

[upsj]

The iteration stopping criterion currently requires that we specify the iteration count using an unsigned integer type, and fails even with integer literals like with_iterations(1) with a hard-to-understand error message based on narrowing conversions in {} brace initialization. With residual norm criteria, the same happens when specifying double values for float parameters.

This PR turns the error into a normal signed-unsigned/lossy conversion conversion warning and hopefully makes it easier to use:

Before:

include/ginkgo/core/stop/iteration.hpp:39:36: error: non-constant-expression cannot be narrowed from type 'int' to 'type' (aka 'unsigned long') in initializer list [-Wc++11-narrowing]
   39 |             this->max_iters = type{value};
      |                                    ^~~~~
examples/simple-solver/simple-solver.cpp:107:47: note: in instantiation of function template specialization 'gko::stop::Iteration::parameters_type::with_max_iters<int &>' requested here
  107 |                 gko::stop::Iteration::build().with_max_iters(max_iters),
      |                                               ^
include/ginkgo/core/stop/iteration.hpp:39:36: note: insert an explicit cast to silence this issue
   39 |             this->max_iters = type{value};
      |                                    ^~~~~
      |                                    static_cast<type>( )

After:

examples/simple-solver/simple-solver.cpp:107:62: warning: implicit conversion changes signedness: 'int' to 'size_type' (aka 'unsigned long') [-Wsign-conversion]
  107 |                 gko::stop::Iteration::build().with_max_iters(max_iters),

[yhmtsai]

could you also share the message old vs new?