Precompile fails with >64 threads

This line fails: https://github.com/JuliaSIMD/CheapThreads.jl/blob/master/src/CheapThreads.jl#L21 if `dynamic_thread_count()` is greater than 64. For example, `(1 << (65 - 1)) - 1 == -1`, which can't be stuffed into a `UInt128`. Casting the leftmost 1 to a `UInt128` at least allow the precompile to work, but I don't know if that is actually correct in the context of the package.