I've created a dozen PRs with changes grouped according to various themes for ease of reviewing. This is the remaining grab bag 😹 where it would be unproductive to decompose further. These changes are all targeted and simple.

• Add braces to an if-statement in ppltasks.cpp.
  • We insist on bracing all control flow.
  • This is compatible with the if-statement being preprocessor-controlled.
• Drop this-> when calling Base::meow().
  • The qualified call is sufficient. (We use this-> when reaching into dependent base classes. These cases are different because we're calling identically-named functions in the base class.)
• Use _STL_INTERNAL_STATIC_ASSERT in invoke().
  • This presumably predated our introduction of test-only static_assert machinery. As "bug in invoke" indicated, if this ever activated, it would indicate an STL implementation bug, not bogus programmer-user inputs.
• Move #include _STL_INTRIN_HEADER to the end so it doesn't interrupt clang-format sorting.
• src/xmtx.hpp doesn't need push-pop defenses.
  • None of our src-internal headers have or need them.
• xmath.hpp and xmtx.hpp should behave like the other src-internal headers and rely on #pragma once instead of idempotency macros.
  • I vaguely recall that xmath.hpp was user-visible a while ago, and xmtx.hpp might have been too, which would explain their different styles.
• Mark _Pass_fn() as _NODISCARD.
  • We call this a lot. Would be a shame if someone discarded its return value.
• _Ryu_exponent can be const.
  • This is a straightforward improvement that doesn't pose concerns for keeping the code in sync with upstream.
• As @strega-nil-ms requested in Implement P2770R0 "Stashing stashing iterators for proper flattening" #3466 (comment) , I've moved the _Can_const_join workaround and _As_lvalue() up, to avoid separating join_view's primary template and specialization.