Document that `take` uses spinning

[tmandry]

Spinning has different performance/energy characteristics than mutexes/condition vars, and it would be good to document that this is what take does.

[hawkw]

Other options include:

• Actually using a condvar here.
  • The reason we're not using one now is that it would only be used in take, and would mean adding a mutex/condvar pair to every slot. This would mean two heap allocations per slot, plus increasing the size of each slot itself by 32 bytes (on linux/x86_64, see https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=5c07e4a6cbf9e3755ff5543a9062d6d9)
  • Using a mutex/condvar pair would also be an obstacle to porting this code to no_std, which is an eventual goal...although we could just make the take method require std.
• Just remove this API entirely in 0.1. I'm not sure how much of a use-case there is for it OTTOMH (tracing-subscriber doesn't use it).

[hawkw]

Another option is that we could consider adding a proper backoff here. We would still be spinning, but could have the thread sleep while spinning rather than just emitting pause instructions. That would be an obstacle to porting to no_std, but this code currently doesn't work on no_std, and we could just feature flag it.

This might be a good compromise between not burning CPU in a tight spin loop and having to allocate a mutex/condvar pair for every slot.