The seqcount latch lock type

I'm glad to see this little trick being useful areas beyond timekeeping. It has some of the benefits of RCU-style synchronization (namely low-latency reads, and reads allowed from NMI context), but unlike RCU, it does not allow read-side to delay updates indefinitely. It's therefore useful in situations where the updater is the kernel, and (at least some) readers are in user-space.

Having its own type is a nice improvement over the original implementation I did for timekeeping.