Compact monotone component

[dannys4]

Similar to how Sven's work parameterizes the map, and using my favorite compile-time tooling, this is a ludicrous addition to MonotoneComponent that changes the parameterization to
$$T(\mathbf{x},y) = \frac{\int\limits_0^y g(\partial_2 f(\mathbf{x},t))\ dt}{\int\limits_0^1 g(\partial_2 f(\mathbf{x},t))\ dt},$$
where $g$ is a rectifier, $f$ is a MultivariateExpansionWorker. Points of note:

• Since this is virtually the same machinery as MonotoneComponent, I just added it as a compile-time feature there. I will (somewhat baselessly?) assert that there's no runtime overhead for "traditional" MonotoneComponent objects (at least, compiled with RELEASE flags).
• As a matter of logistics: note that this doesn't have the $f(\mathbf{x},0)$ in the "traditional" map, since that could escape the hypercube. I wracked my brain trying to allow for this flexibility, but couldn't figure it out.
• For some reason, I see that mixed coefficient jacobians are zero for multi-indices of the form $(k,0)$ and $(k,1)$. I understand the former, but not sure why on the latter? I guess since there's a second derivative, the chain rule ends up eliminating a term, but I'm not sure.
• Since this is compile time, I added the ability to use the usual (non-linearized) bases with the "compactified" basis. I have no idea if this is useful or if I just added a bunch of compile-time cruft; I'm flexible if this should be removed.
• The major overhead added here is the fact that the integral needs to be evaluated twice, and now some derivatives need to get quotient-ruled. I suspect the overhead will be most visible in, e.g., LogDeterminantCoeffGrad, which will become
  $$\nabla_w\log \partial_y T(x,y;w) = \frac{\nabla_w g(\partial_2 f(\mathbf{x},y,\mathbf{w}))}{g(\partial_2 f(\mathbf{x},y,\mathbf{w})} - \frac{\nabla_w \int_0^1 g(\partial_2 f(\mathbf{x},t,\mathbf{w})\ dt}{\int_0^1 g(\partial_2 f(\mathbf{x},t,\mathbf{w})\ dt}$$
  note that this now has integrals, where it previously did not.
• By construction, $T:\ [0,1]^d\times[0,1]\to[0,1]$. As such, I also added a few new bases for the future.
• The only one available via CreateComponent at this point is ShiftedLegendre (which corresponds to BasisTypes::Legendre), which is orthogonal on $U[0,1]$ (note that this is shifted and thus gives a slightly different TTRR than the traditional Legendre).
• I also added SineBasis for the time being, which represents $p_k(t) = sin(2\pi k t)$ but is calculated similar to a TTRR. This can be added, but the current PR is more of a proof of concept.
• While working through the different iterations of this parameterization, I ended up slightly changing the nomenclature for MapFactory's REGISTER_MONO_COMP to work better for future perhaps heterogeneous map types
• This doesn't have bindings. Since we have a developer's main branch, I'm going to make separate PRs for that kind of stuff. I'm also open to merging this into a completely separate branch.