Unify the way callable objects are passed to various API

Currently we have:
```c++
explicit sycl::queue::queue(const sycl::async_handler& asyncHandler, const sycl::property_list& propList = {});
template sycl::event sycl::queue::submit(T cgf);
template sycl::event single_task(const KernelType& kernelFunc);
template void sycl::handler::single_task(const KernelType& kernelFunc)
template void sycl::handler::host_task(T&& hostTaskCallable);
```
Can we optimize more to avoid copies with `T&&` sometimes?
Do we need all these differences? If so we could explain why.
Does the queue keep a copy of the `sycl::async_handler`? A reference?

Compare the callable declarations listed in the issue: queue construction, queue::submit, single_task, handler::single_task, and handler::host_task. First determine the intended ownership and forwarding semantics for each API, then document the rationale and clarify whether the queue copies or references the async handler; done means the differences and copy behavior are specified consistently.

Written by the indexing model from the issue text.