Trackpy

This is primarily a bugfix release. Changes include:
- Fixed bug in `annotate` when specifying a single threshold value.
- Added preliminary (passing) tests of 3D feature-identification capabilities.
- Fixed error in building API reference documentation.
- Removed ephemeral `MANIFEST` file from the distribution.

This release includes some enhancements, bug fixes, and cleanup:
- Enhancements and bug fixes to the new prediction framework
- Ability to use `locate` float-type image data (crucial for compatibility with the new PIMS release, v0.2)
- Faster access to on-disk framewise storage for streaming analysis
- Enhanced multi-color `annotate`

The version contains significant enhancements, including:
- Much-improved feature-finding, merged from Daniel Allan's `mr` project (`feature.py`, replacing `identification.py`) with uncertainty estimation, along with tools for filtering, analyzing, and plotting trajectories
- Prediction framework for tracking particles whose motion is correlated between frames (Nathan Keim)
- KDTree-based linking, merged from Nathan Keim's branch of trackpy, which is 2X faster on typical data
- Numba-accelerated linking and feature-finding, falling back on pure Python if numba is not available
- Features for processing large data sets "out of core" (on disk)
- Access to different linking strategies through keyword arguments (Type `help(link)` or `help(link_df)` for details.)
- Simple, fast way to read and write data in files; easily extensible to formats used by individual research groups
- A set of examples and guides, [provided separately](https://github.com/soft-matter/trackpy-examples)