Test data for event offset smart chunkers

[thorwhalen]

For setting up tests to measure and validate our work for the event offset smart chunkers,
we have several choices:

• real data (with some truth to compare to (and some metric to account for fuzziness of truth)
• synthetic data (some of which as simple as possible so we can include in doctests)
• other functions -- namely more computationally intensive, but accurate ones.

In the latter category, we might consider using convolution and/or cross-correlation.