Looper

Added
- First implementation of pipeline interface 'outputs', so pipeline authors can specify items of interest produced by the pipeline.
- Functions and attributes on `Project` to support "outputs" (`interfaces`, `get_interfaces`, `get_outputs`)

Changed
- Start "compute" --> "compute_packages" transition
- `get_logger` moved to `peppy`

Fixed
- Prevent CLI option duplication in pipeline commands generated
- Make functional CLI spec of particular attribute on which to base selection of a subset of a project's samples ([`peppy` 298](https://github.com/pepkit/peppy/issues/298))

Changed
- Improved documentation
- Improved interaction with `peppy` and `divvy` dependencies

Added
- Implemented `looper rerun` command.
- Support use of custom `resources` in pipeline's `compute` section
- Listen for itemized compute resource specification on command-line with `--resources`
- Support pointing to `Project` config file with folder path rather than full filepath
- Add `selector-attribute` parameter for more generic sample selection.

Changed
- Switched to a Jinja-style templating system for summary output
- Made various UI changes to adapt to `caravel` use.
- Using `attmap` for "attribute-style key-vale store" implementation
- Removed Python 3.4 support.
- UI: change parameter names `in/exclude-samples` to `selector-in/exclude`.

Changed
- `PipelineInterface` now derives from `peppy.AttributeDict`.
- On `PipelineInterface`, iteration over pipelines now is with `iterpipes`.
- Rename `parse_arguments` to `build_parser`, which returns `argparse.ArgumentParser` object
- Integers in HTML reports are made more human-readable by including commas.
- Column headers in HTML reports are now stricly for sorting; there's a separate list for plottable columns.
- More informative error messages
- HTML samples list is fully populated.
- Existence of an object lacking an anchor image is no longer problematic for `summarize`.
- Basic package test in Python 3 now succeeds: `python3 setup.py test`.

Changed
- Fixed bugs with `looper summarize` when no summarizers were present
- Added CLI flag to force `looper destroy` for programmatic access
- Fixed a bug for samples with duplicate names
- Added new display features (graphs, table display) for HTML summary output.

Changed
- Fixed several bugs with `looper summarize` that caused failure on edge cases.