Remove `vcat` and `hcat` in favour of `ncat`

[c42f]

vcat and hcat seem a bit redundant now that we have generalized concatenation. Maybe we should we always just emit ncat and only convert to vcat/hcat in Expr conversion.

[c42f]

Looking into this a bit, it becomes less clear we should do this because of row vs column-major considerations.

With spaces we have row major syntax so [a b; c d] is

   1:1  │[vcat]
   1:2  │  [row]
   1:2  │    a
   1:4  │    b
   1:7  │  [row]
   1:7  │    c
   1:9  │    d

With semicolons we have column major syntax so [a ; b ;; c ; d] is

   1:1  │[ncat-2]
   1:2  │  [nrow-1]
   1:2  │    a
   1:6  │    b
   1:10 │  [nrow-1]
   1:10 │    c
   1:14 │    d

In normal ncat syntax, the dimensions are strictly nested with highest dimension outermost because more semicolons have lower precedence than less.

To represent row major syntax in ncat, we'd use a tree where the first and second dimension are nested in the opposite way:

   1:1  │[ncat-1]
   1:2  │  [nrow-2]
   1:2  │    a
   1:6  │    b
   1:10 │  [nrow-2]
   1:10 │    c
   1:14 │    d

So it's quite possible to do this with ncat.

However, we diverge further from the surface syntax by abstracting to ncat like this and users would then need to test ncat nodes carefully for row-major layout. So overall this doesn't feel like the win I thought it would be.