newTimeout is not safe on `32`-bit archs

[coot]

As noted by @njd42 :

On a 32 bit machine with a timeout more than ~35min in the future defaultRegisterDelay would be invoked which, in turn, makes use of newTimeout. This calls GHC.registerTimeout where the intrinsic 32 bit constraint rears its ugly head again.

[coot]

We haven't fixed newTimeout bug in the pr #57 addresses safety of MonadTimer primitives in the si-timers package.