DEMetropolis: Warn on too small population

[michaelosthege]

As explained by ter Braak (2006), the number of chains should be larger than the number of dimensions:

What happens if N ≤ d? Because N points lie in an N − 1
dimensional space, all proposals (2) will lie in this reduced
space when e = 0. Therefore convergence of DE-MC would
rely on e, which would take a long time if its variance is small.
To stress that convergence does not depend on the unbounded
support of e, we actually used e ∼ Uniform[−b, b]^d with
b = 10⁻⁴ in all computations (Fig. 2). In the next section
the effect of N on the efficiency of DE-MC is studied via
simulation for N > d.

Therefore, we should warn the user about potentially inefficient sampling when n_chains <= n_dims