Address unique_span issues from cr.SX

[eyalroz]

I've asked for a code review of my unique_span class, on codereviews.stackexchange.com, and got one.

Issues brought up:

1. Can simplify swap()
2. No need to delete copy ctors other than those C++ might itself generate
3. Assignment to the base-class is long-winded
4. We should be able to convert from a span of T to a span of const T.
5. We don't know that release() is noexcept... so making any of our [move] constructors noexcept is a leap of faith.
6. Deleter is not stored
7. Factory method is not allocator-aware
8. Factory method requires default-constructible element type
9. No Tests

Well, my decision about all of those:

1. Do it.
2. It's a form of documentation-via-code... plus I'm not 100% sure this can't happen, e.g. via type conversion which isr's b
3. Neat, let's do it. But - assigning {} is a bit confusing for the reader, IMHO, so let's keep that part explicit.
4. Do it.
5. Do it - drop the noexcept.
6. Mulling over whether I want to support a stateful deleter. For now - not doing it.
   edit: see #678 .
7. Not right now.
8. Not right now.
9. Ah, tests... we all want tests.

[eyalroz]

Some issues can't be fully addressed due to the use of the "poor man's span" with earlier C++ versions, as the base class.