Remove local (in-group) atomic usage from `__parallel_find_or`

[SergeyKopienko]

In this PR we remove local (in-group) atomic usage from __parallel_find_or implementation:

• __found_local now is local variable (for each item);
• barriers doesn't required anymore;
• we updates the state of __found_local within one group through __dpl_sycl::__reduce_over_group operation :

                    __found_local = __dpl_sycl::__reduce_over_group(__item_id.get_group(), __found_local,
                                                                    typename _BrickTag::FoundLocalReduceOp{});

where FoundLocalReduceOp is :

• __dpl_sycl::__minimum (for the __parallel_find_forward_tag);
• __dpl_sycl::__maximum (for the __parallel_find_backward_tag);
• __dpl_sycl::__plus (for the __parallel_or_tag).