Improved bootstrap filter

[zsunberg]

We should probably make an improved bootstrap filter that has the following changes:

1. We should only resample periodically (see Add resampling criteria #24 Add ESS resampler to avoid resampling every iteration #67 )
2. Because of periodic resampling above, we should return a weighted particle belief rather than unweighted
3. We should have an additional function for particle reinvigoration that is separate from resampling, because that is the most common domain-specific thing that people need to do.
4. predict, reweight, and resample should just be functions (which can be implemented as functors)