TDM schema hardening

[brettviren]

This is a low-priority wishlist todo reminder for when someone is bored.

The tensor data model schema and I/O methods should be "hardened". Right now we hand-craft I/O converters between higher level WCT structures (ICluster, PointCloud::Dataset, etc) and file representations. It requires careful human attention at every point to develop and maintain.

Better would be to add a subsystem that goes something like this:

1. Formally represent the TDM schema with moo schema.
2. Develop a C++ moo template file operating on a TDM moo schema object that renders to a C++ struct representing that TDM object.
3. Develop a C++ moo template file operating on a TDM moo schema object that generates I/O methods converting between the above struct and a file representation. One template targeting the WCT json+numpy+archive format and one targeting hdf5.
4. Develop a C++ moo template operating on a collection of all TDM moo schema objects that generates the necessary dynamic-static type conversion needed to serialize a file of multiple TDM records of differing types.

See #267 for recent work in this area and in particular not the use of the located index which is required to speed up I/O operations. This index might live as part of the generated code in 3 or 4.