Converging toward P0896R3 (algorithms and special memory algorithms)

[cjdb]

This commit moves cmcstl2 closer toward what will be in P0896R3 for San Diego.

Changes

• Return types more closely resemble what's in R2. I created exposition-only base structs instead of creating aliases to see how well that would work with the library as a whole.
• Removed a lot of dependency on tagged.hpp in the wake of named return types.
• Most algorithms are now implemented as functors, and are not lookupable by ADL, as required by [specialized.algorithms]/3.1 and [algorithms.requirements]/2.
  • Many details now cannot be looked up at all: they're private static members.
  • This potentially eliminates the need for STL2_WORKAROUND_GCC_79591, as this bug doesn't seem to surface when a using declaration is made in a class definition.
• Algorithm interfaces resemble what is present in ericniebler/stl2@19b9f15.
  • Optimisations have been made private static members that are delegated to by the public interface. The intention is to significantly reduce the noise given by diagnostics for people unaware that algorithms can be optimised for certain concepts.
• Some, but not all, of Rng has been renamed to R.
• safe_iterator_t now disallows rvalues: as @CaseyCarter predicted on Slack, this is problematic when passing range adapters into algorithms.

What's missing?

• I haven't looked at shuffle yet, but I imagine it'll follow the same formula. I can add it to this branch, or I can add it to a different one.

Things that must happen before merging with master:

• The test case for find_end is uncommitted (and incomplete) on my work computer, so I've disabled compiling that test to get this PR up early. It'll need to be committed before merging, as there are very likely issues with find_end that I haven't had the opportunity to pick up on yet.
• I'd like to rebase and squash this. The commits are useless fluff, mostly so I could change between home and work computers.

Issues this might close

• constexpr extensions #17 constexpr extensions
• Sweep the codebase for components out-of-spec to the C++20 WP #41 Sweep the codebase for components out-of-spec to the Ranges TS WP
• Re-tighten constraints on algorithms #70 Re-tighten constraints on algorithms
• Whither deprecated algorithm overloads? #188 Whither deprecated algorithm overloads?

Issues this pull request opens

• nth_element uses goto #189 nth_element uses goto.

[cjdb]

Conflicts with master way too much, and it'll be easier to redo this in piecemeal chunks anyway.