Add a fail-fast retry strategy, and document them.

Since we currently only have a global retry strategy (per lcb_INSTANCE), some clients (say the python or node client) may want to switch it off – and never retry.  Or maybe other things as well.  But for sure a fail-fast that just doesn't retry is something that may be wanted.  Though probably this is more useful later when we will be able to specify a retry strategy for a request, lets add the fail-fast (never retry), so clients can at least switch the best_effort strategy on/off.

While at it we can add a bit of documentation for `lcb_retry_strategy` as well.