Atropos

* Add FASTA output for detect command, and enable json, yaml, and pickle output for all commands.

* Major update to the documentation.
* Fixed error messages in multi-threaded mode.
* Fixed bug when generating reports for runs involving error correction.

* Exposed option to set PRNG seed when subsampling reads.
* Fixed issue 14: 'detect' and 'error' commands were broken. This involved rewriting those commands to use the same pipeline and reporting frameworks as the 'trim' and 'qc' commands.

* Updated Dockerfile to use smaller, Alpine-based image.
* Added Docker image for v1.1.2 to Docker Hub.
* Updated Travis config to automatically build Docker images for each release.
* Ported over improvements to adapter parsing (635eea9) from Cutadapt.
* Fixed 12: tqdm progress bar not working.
* Fixed 13: unnecessary differences in summary output between Cutadapt and Atropos.

* New 'qc' command computes read-level statistics.
* The 'trim' command can also compute read-level statistic pre- and/or post-trimming using the new '--stats' option.
* Major refactoring and improvement of reporting:
* Text report now has data lined up in columns.
* Reports can also be generated in JSON, YAML, and pickle formats.
* Added optional dependency on jinja2, which enables generating reports using templates (including user-defined).
* Major internal code reorganization.
* Static code analysis (pylint).
* Switched to pytest for testing.
* Command-specific help will now show with 'atropos <command>' or 'atropos <command> -h'
* Fixed adapter masking in InsertAligner (issue 7; thanks lllaaa).
* Added developer/contributor documentation and guidelines.
* Implemented Atropos module for MultiQC, which reads reports in JSON format. This is currently available [here](http://github.com/jdidion/atropos-multiqc) and will hopefully soon be merged into MultiQC.
* Ported some recent enhancments over from Cutadapt.

* Identified a subtle bug having to do with insufficient memory in multi-threaded mode. The main thread appears to hang waiting for the next read from the input file. This appears to occur only under a strictly-regulated memory cap such as on cluster environment. This bug is not fixed, but I added the following:
* Set the default batch size based on the queue sizes
* Warn the user if their combination of batch and queue sizes might lead to excessive memory usage.
* Bug fixes