Allow JOB_TASKS_PLAN to be configurable per publication

[jpmckinney]

For example, extremely large datasets (like ProZorro Ukraine, and previously a few Digiwhist publications) will probably not be flattened in a reasonable amount of time.

If large publications only have one release per contracting process, we can replace the process step by either:

• Running an alternative worker to Kingfisher Process, which reformats the packages from Kingfisher Collect as compiled releases.
• Adding an option in Kingfisher Collect, which causes AddPackageMiddleware and KingfisherProcessAPI2 to be skipped, and causes a new RemovePackageMiddleware to be run. (This is probably faster, if not overly complicated.)

We would need a new exporter that indexes files on disk, and then writes the monthly and yearly files.

Both options will create a lot of files, which requires open-contracting/kingfisher-collect#740 to be resolved, otherwise the performance will be awful.

Before adding ProZorro, we should also complete #291.

So:

• Make JOB_TASKS_PLAN configurable per publication (simplest implementation might be a "very large dataset" checkbox that causes an alternative plan to be used). At minimum, this should skip flattening.
• Resolve Some spiders create millions of files in a directory kingfisher-collect#740
• Resolve Replace Pelican with Cardinal in pipeline (and make detailed coverage available) #291
• Decide on and implement the alternatives to the process and exporter steps