Fix tolerance used for floating point comparison (Fixes #851)

[andrewdnolan]

Per @trhille's experimenting, there are problems setting the Dirichlet BC's for resolutions $\leq$ 1000(m). This PR fixes #851 by using an absolute tolerance in all floating point comparisons of 5% of the appropriate cell spacing quantity.

[andrewdnolan]

Testing:

Current implementation

Here's the quantities of interest from the HEAD of main:

As @trhille pointed out there are too many cells along the North boundary marked as Dirchlet, same problem applies to areaCell. dvEdge is a bit more complicated I'll explain that latter.

Using a 5% of cell spacing absolute tolerance

Here's the quantities of interest produced with this branch:

There are two improvements:

1. The culling is now correct in that the top most cell center corresponds to exactly to 80e3 (c.f. before).
2. The masking of dirichletVelocityMask and areaCelll now only affect the top most cell.

[andrewdnolan]

Testing (Cont.)

So about the dvEdge field. Unfortunately the way I construct edge patches in mosaic is too simple to deal with these slightly complicated boundary edges. So, for now I've got to go to paraview (for now).

Here's an example of the same dvEdge field for a 1000m mesh, where the edges above the boundary (80e3) have been set to zero and the edges at the boundary have been set to half their original distance. Which is needed to match the half area of the cells along the boundary (see above).

[trhille]

Thanks @andrewdnolan! I tested 1000m and 500m mesh creation and it now works as expected.