Enable default behavior for pmin, predux_min, etc

Reference issue

What does this implement/fix?

Currently, the default implementation for predux_min<PropagateFast> (and max) defaults to a scalar loop. This MR forwards predux_min<PropagateFast, Packet> to predux_min<Packet>, which is available on most platforms. If the packet type is integral, all propagation types are forwarded to predux_min<Packet>. Similarly for pmin<NaNPropagation>.

In summary:

Floating point types:

Requested	Default
op<PropagateFast, Packet>	op
op<PropagateNaN, Packet?	specialized
op<PropagateNumbers, Packet>	specialized

Integral types:

Requested	Default
op<PropagateFast, Packet>	op
op<PropagateNaN, Packet>	op
op<PropagateNumbers, Packet>	op

This deletes the PropagateFast specializations for the intel platforms as they are redundant.

Additional information