Accelerate minimum(A; dims = 1) for cartesian indexed cases.

[N5N3]

Our mapreduce use mapreduce_impl for linear-indexable inputs to accelerate min/max related calls, which is critical to Float cases especially when dim1 is included in reduction.
But the optimization is missing for cartesian cases.

This PR try to fuse more mapreduce_impl for free speed:
minimum Benchmark (combined with #43604, which widen the optimizaiton to much shorten length):

using Random, BenchmarkTools                   
Random.seed!(0);
a = view(rand(128,128),1:128,1:128);# PR 43604      This PR
@btime minimum($a, dims = 1);       #28.100 μs  --> 9.400 μs 
@btime minimum($a, dims = 2);       #4.071 μs   --> 4.086 μs 
@btime minimum($a, dims = (1,2));   #26.700 μs  --> 9.700 μs
@btime minimum($a);                 #30.400 μs  --> 9.100 μs