P2897 (aligned_accessor): R1

[mhoemmen]

P2897R1, to be submitted 2023/10/15

Make wording changes requested by LEWG review on 2023/10/10

• [x] Change gcd converting constructor Constraint to a Mandate
• [x] Add Example in the wording section that uses is_sufficiently_aligned to check the pointer overalignment precondition
• [x] Add Example in the wording section that uses aligned_alloc to create an overaligned allocation, to show that aligned_accessor exists as part of a system
• [x] Add an explicit constructor from default_accessor, so that users can type aligned_mdspan y(x) instead of aligned_mdspan y(x.data_handle(), x.mapping(), aligned_accessor<element_type, byte_alignment>(x.accessor())).

Add design discussion based on LEWG review on 2023/10/10

• [x] Explain why we do not include an aligned_mdspan alias

Make other wording changes

• [x] Add to aligned_accessor's Mandates that byte_alignment >= alignof(ElementType) is true. This prevents construction of an invalid aligned_accessor object.

Fixes #410 .

[mhoemmen]

@crtrott wrote:

The examples are probably more than LWG would want.

It's OK -- LEWG specifically asked for them. We can always move them into the non-wording part of the paper later.