atleast_2d before super-post-init

[gdmcbain]

Coercion of ‘scalar’ one-dimensional coordinates to 1-vectors needs to happen before Mesh.__post_init__; otherwise the trace of a two-dimensional basis with a ‘scalar’ one-dimensional projection hits #686.

[kinnala]

Why no test caught this?

[gdmcbain]

Indeed. Let me fix that.

[gdmcbain]

The only test always passes an (n − 1)-dimensional projection.

scikit-fem/tests/test_basis.py

Line 250 in e7196f6

nbasis, y = basis.trace(xfun, lambda p: p[0:(p.shape[0] - 1)], target_elem=e2)

(It did cross my mind yesterday that maybe the points passed to construct a one-dimensional mesh should have ndim==2 and shape[0]==1 but that is a bit too strict, isn't it. People do want to be able to pass just ndim==1 arrays of just the x-coordinate, or even lists #663, and it would seem to be pythonic to accept them.)

[gdmcbain]

This works (fails on master, passes here).

The pytest.mark.parametrize is getting verbose; maybe it could be shortened with some generation or itertools, but I've left that for now.

[gdmcbain]

Maybe a full factorial test against #686 isn't required, just a single test with flat doflocs; I doubt it interacts with the other factors.