Always use environment variables from add time + minor changes

[n-gao]

Functionality change:

• Environment variables will now be stored in the MongoDB at add time and used when submitting a job. This allows you to dispatch jobs from a different environment than the one you used for adding.

Logging changes:

• We now display the correct number of files we deleted and do not count duplicates.
• We don't throw random exceptions when a job cannot claim an experiment but fail gracefully.

Performance changes:

• We use regex to figure out matches for interpolation keys (seml add is quite a bit faster)
• We avoid resolving the description twice (faster seml add)
• deleting files from the MongoDB is now running in parallel (seml delete is now pretty much instant)
• Files are uploaded with multprocessing (faster seml add)

[n-gao]

@dfuchsgruber, the last commit is an important fix. Seml would call scancel -t PENDING [job_id, ...] if you call seml cancel. However, if the collection is empty, it would just call scancel -t PENDING resulting in killing all pending jobs of a user. In case the user is a slurm admin, it will kill all pending jobs on the cluster.

We should merge this asap and make a new release!