kokkos(ext): support when_all

[romintomasetti]

The problem is that Kokkos dispatches asynchronous work.

But while e.g. exec::static_thread_pool will simply call set_value_t(next_receiver) from the thread that performed the work, such that the next_receiver only starts its work once the current receiver has finished, it cannot be done like that with Kokkos dispatch.

My first approach was to customize schedule_from to properly synchronize when needed. But it seems it will not be called e.g. from within a when_all:

... | stdexec::when_all(fork | stdexec::continues_on(exec.get_scheduler()) | ...);

So I tried to get some inspiration from https://github.com/NVIDIA/stdexec/blob/485160802ee5ca42ca4915e3a2330579efae4ea3/include/nvexec/stream/common.cuh#L621-L629, that seems to do a continuation under some circumstances. Not sure it is the best way though.

[maartenarnst]

Could be worth taking a look at:

• https://github.com/NVIDIA/cccl/blob/61ea9e8b4eac0f877f28c61551c01a3b5642bd40/cudax/include/cuda/experimental/__execution/stream/schedule_from.cuh#L48-L66

[maartenarnst]

• https://github.com/NVIDIA/cccl/blob/b3a64ca52c23cf4baac80b064c2171e286242f71/cudax/include/cuda/experimental/__execution/stream/bulk.cuh#L71C7-L71C48