2D: do not return points outside the bounds, closes #181

[basnijholt]

@andrenmelo, this is the (a) solution.

@python-adaptive/core what do you think?

[akhmerov]

Here the numerical errors don't accumulate over time, and therefore we may simply choose a relative tolerance that is merely much higher than the double precision (e.g. 1e-13), and not go over the complete interpolation each time.

Also as implemented, there is no guarantee that all the points that are selected are within the triangulation if e.g. the midpoint on the edge of the rectangle is outermost and the top or bottom are ε away from it.

[basnijholt]

Also as implemented, there is no guarantee that all the points that are selected are within the triangulation if e.g. the midpoint on the edge of the rectangle is outermost and the top or bottom are ε away from it.

I think that the rounding of the same number should always happen the same, so I don't think this could occur.

Here the numerical errors don't accumulate over time, and therefore we may simply choose a relative tolerance that is merely much higher than the double precision (e.g. 1e-13), and not go over the complete interpolation each time.

So you just mean doing?

eps=1e-13x=y=np.linspace(-0.5+eps, 0.5-eps, n)

[akhmerov]

LGTM