tmc::fork_group can be awaited multiple times

[tzcnt]

Previously tmc::fork_group was an rvalue-only awaitable, like most other TMC awaitables. This restriction makes sense for awaitables that initiate work and retrieve results in the same operation (since initiating work is not idempotent). However, fork_group separates the steps of initiating work and awaiting it. So we can make it an lvalue awaitable which can be awaited multiple times to retrieve the same result again. Or the user can still cast to rvalue to move-from the result. If the user wants to dispatch another group of work later, they can call reset() before doing so.

This allows it to replace tmc::manual_reset_event in some places, with a caveat: awaiting manual_reset_event is thread-safe as it has a mechanism to register multiple awaiters, but fork_group does not. So only a single thread should await fork_group at any given time.

Supported usages:
co_await fg; // returns ResultArray&
co_await std::move(fg); // returns ResultArray&&
std::move(co_await fg); // returns ResultArray&&