Add docstring generation for Python wrapper

[p-zach]

Adds the capability to extract Doxygen documentation from the C++ source code and insert these docs into the Python bindings. This means that the documentation is then available in the help() function. The functionality can be seen in these examples:

>>> import gtsam
>>> help(gtsam.Pose3.transformTo)
Help on instancemethod in module gtsam.gtsam:

transformTo(...)
    transformTo(*args, **kwargs)
    Overloaded function.

    1. transformTo(self: gtsam.gtsam.Pose3, point: numpy.ndarray[numpy.float64[3, 1]]) -> numpy.ndarray[numpy.float64[3, 1]]

    takes point in world coordinates and transforms it to Pose coordinates
    point: point in world coordinates
    Returns: point in Pose coordinates

    2. transformTo(self: gtsam.gtsam.Pose3, point: numpy.ndarray[numpy.float64[3, 1]], Hself: numpy.ndarray[numpy.float64[m, n], flags.writeable, flags.f_contiguous], Hpoint: numpy.ndarray[numpy.float64[m, n], flags.writeable, flags.f_contiguous]) -> numpy.ndarray[numpy.float64[3, 1]]

    takes point in world coordinates and transforms it to Pose coordinates
    point: point in world coordinates
    Hself: optional 3*6 Jacobian wrpt this pose
    Hpoint: optional 3*3 Jacobian wrpt point
    Returns: point in Pose coordinates

    3. transformTo(self: gtsam.gtsam.Pose3, points: numpy.ndarray[numpy.float64[m, n]]) -> numpy.ndarray[numpy.float64[m, n]]

    transform many points in world coordinates and transform to Pose.
    points: 3*N matrix in world coordinates
    Returns: points in Pose coordinates, as 3*N Matrix

>>> help(gtsam.Pose3.AdjointMap)
Help on instancemethod in module gtsam.gtsam:

AdjointMap(...)
    AdjointMap(self: gtsam.gtsam.Pose3) -> numpy.ndarray[numpy.float64[6, 6]]

    Calculate Adjoint map, transforming a twist in this pose's (i.e, body) frame to the world spatial frame Ad_pose is 6*6 matrix that when applied to twist xi $ [R_x,R_y,R_z,T_x,T_y,T_z] $, returns Ad_pose(xi) 

>>> help(gtsam.Cal3_S2.calibrate)
Help on instancemethod in module gtsam.gtsam:

calibrate(...)
    calibrate(*args, **kwargs)
    Overloaded function.

    1. calibrate(self: gtsam.gtsam.Cal3_S2, p: numpy.ndarray[numpy.float64[2, 1]]) -> numpy.ndarray[numpy.float64[2, 1]]

    Convert image coordinates uv to intrinsic coordinates xy.
    p: point in image coordinates
    Returns: point in intrinsic coordinates

    2. calibrate(self: gtsam.gtsam.Cal3_S2, p: numpy.ndarray[numpy.float64[2, 1]], Dcal: numpy.ndarray[numpy.float64[m, n], flags.writeable, flags.f_contiguous], Dp: numpy.ndarray[numpy.float64[m, n], flags.writeable, flags.f_contiguous]) -> numpy.ndarray[numpy.float64[2, 1]]

    Convert image coordinates uv to intrinsic coordinates xy.
    p: point in image coordinates
    Dcal: optional 2*5 Jacobian wrpt
    Dp: optional 2*2 Jacobian wrpt intrinsic coordinates
    Returns: point in intrinsic coordinates

To generate these Python docstrings, follow these steps:

1. Build the Doxygen XML. It's better to use the built Doxyfile in build/<build-name>/doc, but it might still work if you use the doc/Doxyfile.in (if, for example, you haven't built GTSAM yet and don't want to build it just to compile the Doxyfile).
2. Set the new Cmake variable GTSAM_BUILD_PYTHON_DOCS_FROM_XML to ON.
3. Set the new Cmake path variable GTSAM_PYTHON_DOCS_XML_SOURCE to the path to the xml folder generated by Doxygen.
4. Build GTSAM with Python and pip install . it.

The parser is fairly robust but not perfect. I encourage anybody to test it out on functions that have well-formatted documentation in the source code so that we can improve it (naturally, the parser can only work with what's in the source--bad or poorly formatted source docs --> bad or poorly formatted Python docs).

To build these docstrings automatically with the pypi package that @yambati03 is working on, the build process will need to be amended so that Doxygen XML is generated before GTSAM Python is built. It takes about 3 minutes to generate the Doxygen files on my machine, and that can probably be shortened if necessary by editing the Doxyfile to not build unnecessary files.

Lastly--this PR edits wrap. Does that mean the changes need to be duplicated in the wrap repo? Not sure what the practice is there.

[dellaert]

Cool! Some issues:

• indeed, wrap is pulled in from a different repo. Maybe @varunagrawal can comment on correct way to change wrap and get it back into GTSAM.
• that also affects the flag names: wrap is not GTSAM, so cmake variables there have a different naming convention I think
• Finally, is it possible to not need GTSAM_PYTHON_DOCS_XML_SOURCE?