Un-deprecate NoiseModelFactor1-6 and move X1 and key1-6 shortcuts to NoiseModelFactorN

[gchenfc]

Following the discussion from #1350, this PR replaces #1350 and undoes some changes from #1344:

1. Moves X1-X6 aliases into NoiseModelFactorN (not deprecated)
2. Un-deprecates key1()-key6() shortcuts in NoiseModelFactorN
   • and we can also undo some particularly ugly changes needed in Replace NoiseModelFactor1 in with NoiseModelFactorN in pre-made factors #1344 due to templates on dependent types. The code should still work fine without undo-ing, but I think it's easier to read this way.
3. #define NoiseModelFactor1 NoiseModelFactorN. Now that NoiseModelFactorN's API is a superset of NoiseModelFactor1-6, we can simply define it so.

As for 3, template <T> using NoiseModelFactor1 = NoiseModelFactorN<T> doesn't work because some detail regarding templates and class name injection. See #947 (comment)

Primary change is in NonlinearFactor.h. The rest of the files are just undo-ing some changes from #1344 and un-disabling deprecation warnings in testNonlinearFactor.h.

---

Copy-pasting pros and cons from previous discussion for reference:

Pros:

• Makes migrating from NoiseModelFactor1-6 to NoiseModelFactorN way easier, since no calling-side code has to change at all. Just find-all-replace : public NoiseModelFactor[1-6] -> : public NoiseModelFactorN
• keep things shorter/more readable. MyFactor::X3 is (subjectively) cleaner than MyFactor::ValueType<3>.
• Still extensible: in the event someone needs e.g. X7, then they can just use ValueType<7>.
• There's one particular case that makes ValueType<1> annoying to use: when you are defining a templated factor and you want to use ValueType<1> inside the definition of the templated factor, you need to do typename MyFactor::template ValueType<1> instead of just ValueType<1>. So in these cases, it's much nicer to just use X1.

Cons:

• clutters up implementation of NoiseModelFactorN making it slightly difficult to read (now inherits from detail::NoiseModleFactorAliases<ValueTypes>).
• since we would do #define NoiseModelFactor1 NoiseModelFactorN, some compilers may not support deprecating the old classes, and generally #define is kind of bad practice (?)

[dellaert]

Oh, wow, this is much better !!!!

[dellaert]

Feel free to merge when CI passes