Precise axes in generic_mattrimul!

[jishnub]

This changes e.g. axes(B,1) to axes(A,1) in an expression like

for j in axes(B,2)
-    for i in axes(B,1)
+    for i in axes(A,1)
        Cij = (unit ? oA : A[i,i]) * B[i,j]
        for k in i + 1:lastindex(B,1)
            Cij += A[i,k] * B[k,j]

While we obtain the correct result either way, since, firstly, triangular matrices are square, and secondly, the axes being multiplied must match, it's best to be precise here.