Constrain completion CPOs to non-const rvalue receivers

[emmett2020]

The standard requires calls to set_value, set_error, and set_stopped to be ill-formed when the receiver is an lvalue or a const rvalue. The current implementations only check member callability and do not enforce these restrictions. This change adds the missing constraints and corresponding tests.

The stricter constraints expose existing lvalue completion calls in parallel_scheduler and merge_each. The scheduler now passes its receiver as an rvalue. In merge_each, the next-sequence operation now stores a decayed receiver type to allow rvalue completion and avoid retaining a dangling reference:

auto subscribe(auto receiver) &&
{
  return ex::connect(next_sender, receiver);
  // Before: operation stores R&; completion cpo receives an lvalue.
  // After:  operation stores R; completion cpo receives an rvalue.
} // Storing R& also leaves a dangling reference to this local receiver.

This MR also includes 4 unrelated formatting-only line changes required to pass the CI clang-format 21 check.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.