let schemas specify their own toJSON

[hjdivad]

consider case of saving an existing frame and only pull from it at serialization time

[hjdivad]

quite possibly allow toJSON to return virtual nodes (eg as an array [0, 100], [20, 80], [40, 60])

[hjdivad]

think about how to fit this in sensibly with #16

[wycats]

I wanted to leave some notes explaining what requirements I had in mind. I would like to be able to:

• put a lot of high-fidelity markers in Glimmer that I leave in unconditionally.
• control the code that runs when the markers are reached, so that it's possible to make them do nothing at all.
• allow these markers to represent starting and ending events, and be able to track the time between them, but while still maintaining the ability to turn these markers into noops.
• to track nesting (before:render -> before:render -> after:render -> after:render has two conceptual "frames") without being forced to reify Heimdall context nodes; in the case of Glimmer, I already have perfectly suitable frames lying around that I can add the context information to.
• for this aggregate information to be usable to general-purpose aggregation tools that want to present things like "ordered self time" or a tree visualization.

I am willing to accept a requirement that time must be tracked using Heimdall's preferred clock (performance.now in Chrome) if it is to be included in aggregation.