This commit sets up the basic SBT subproject configuration and
moves the core algebra type classes into the cats.kernel package.

Algebra doesn't currently provide generic semigroups for
number types, so this commit doesn't have any of those
either. Cats is already working around this so that
shouldn't be a big change.

There are a few things here:

 1. Many of algebra's interesting laws apply to rings/lattices.
 2. We introduce a (laws-only) dependency on catalysts-platform.
 3. Unlike Cats, we'll run the tests in kernel-laws' tests.

This commit ports over the law tests which make sense in
the context of cats-kernel. For some reason the tests aren't
being run right now, but at least they compile.

After this commit, the intention is that none of the algebra
types are used. The tests pass and things seem to work. Next
step is to actually remove the algebra dependency.

After clean, validate passed. I think this branch is good to go.

This commit fixes some problems which better testing
exposed. The biggest was that when we moved from the
"sparse vector" interpretation of maps to the "strict"
interpretation some of our instances became unlawful.
We can't define a group directly on maps because we
can't enforce the equivalence of "absent key" and
"present key with empty value".

This is an optimization that we've known about for awhile and
that Oscar suggested should be added to some of the instances
for Eq, PartialOrder, and Order (those where the
cost of doing the check can be linear).

This does the following:

 1. Fixes a lot of return types.
 2. Removes cats.kernel.std.array
 3. Removes the unused cats.kernel.Priority
 4. Improves some comments
 5. Makes a small efficiency improvement for Monoid[Map[K, V]]

This fixes the merge conflicts and also re-removes algebra.

This commit also adds some "missing" function traits
(as abstract classes) and replaces an instance of
c.isNaN with isNaN(c) for performance.