Implement support for S3 polling as alternative to lambdas

[jrask]

Lambdas can only run for 5 minutes, can not read old files, only 3 retries... etc.

Implement support for polling S3.

• Conf max concurrent requests through pipeline
• Must be stateless (sort of keep last in memory but require rename or move of files). To complicated to keep track polled files.
• NotOlderThan(int, TimeUnit) to make it possible to not poll old files.
• Filter, suffix, prefix.

Fill with more as we go.