Replace point constants with static class members

[sparr]

Summary

None

Purpose of change

Currently whenever someone needs one of the point constants (zero, min, max, invalid, north, north_east, etc) for a coordinate type that hasn't been needed before, they add it to coordinate_constants.h. This file has accumulated over 100 such constants, in a hodge podge with no clear pattern to which constants exist and which constants don't and when it's appropriate to add a new one. This can make the codebase confusing to someone who sees tripoint_abs_omt_foo and wonders why tripoint_rel_ms_foo doesn't exist.

Describe the solution

I have moved the constants into static members of the point and tripoint classes, then exposed them in coord_point_ob. Now every tripoint_foo_bar type has ::zero, ::min, ::max, etc. The relative types have ::north, ::north_east, etc. The relative 3d types have ::above and ::below.

These are implemented such that the conditional static member is private and named _north and an unconditional untemplated public member named north is a reference to it, which was necessary to avoid needing an empty template parameter list at every usage (e.g. ::north<>). This means that IDEs may autocomplete ::north on types that don't have ::_north, which will produce a compiler failure.

I also added a helper function to check whether a [tri]point equals the ::invalid constant, which was the original goal of this effort and led me down the rabbit hole of making point and tripoint behave more compatibly.

I've attempted to update every usage, including various places that were using _min but should have been using _invalid (which didn't exist in some cases) and are now using ::invalid.

I updated the tests that use relative constants with absolute coordinate types, to instead create an absolute coordinate from a relative underlying point.

I described these constants in POINTS_COORDINATES.md.

Describe alternatives you've considered

Exposing the constants on the coordinate types could be done in coord_point_base, but that felt too bloated.

Testing

Automated tests pass. I am playing with this change.

Additional context

I'm not sure how best to break this up into multiple PRs. I could put the old constants back, change just a few files to the new constants, PR that, then continue until they are all changed?

I expect this to fail on MSVC, and will address that when it happens.

[Procyonae]

Adding these to POINTS_COORDINATES.md wouldn't hurt

[CLIDragon]

Given the changes are trivial and easily verified I feel like it should be okay even as a massive line diff.

[CLIDragon]

Should this be const?

[sparr]

This should really be implementing point==tripoint and using the existing eight_horizontal_neighbors instead of defining a new static array, but I'm not going to throw a functional change like that into a PR that's already this big. I'll add const here and come back to removing this on the next pass.

[PatrikLundell]

This looks like a good change.

There was actually a sort of intention behind the set of constants that are defined and the ones that weren't, and that was that only it only made sense to have offsets for the relative types, with zero being used fairly frequently for absolute ones as well. However, it's only mildly detrimental to define everything for all types (detrimental, because it makes it easier to screw up and use an absolute offset constant when you really should have used a relative type, and similar errors).