The database behind Cloudant, CouchDB uses an append-only file for it's data. To restore already used up space, a compaction must happen. When this happens is up to the database maintainer.
Cloudant hosts databases with a lot of different cloud hosting providers including Amazon, Rackspace, SoftLayer and Microsoft Azure. This way customers can choose where their database is hosted.
Since CouchDB is considered an AP (Available, Partition-Tolerant database management system), it is not really consistent (not all clients can have the same view of the data consistently) and the only way to achieve some "eventual consistency" is through replication and verification of data.
When a new record is written, it can trigger one or many AWS Lambda functions. With Lambda functions in Java, JavaScript, and Python and the other con of "Easy integration with other Amazon services", Lambda functions may be all you need to process the events. This is particularly useful for creating projections (often as other DynamoDB tables) and other read models in a CQRS pattern.
Data hosted on DynamoDB is automatically replicated across multiple AWS availability zones, this way the data is protected from any malfunction that may cause loss and it also ensures high uptime.
Since it's part of AWS, like with other services that Amazon provides, it's very easy to integrate DynamoDB with any other AWS service that you may be using.
DynamoDB is just a general purpose NoSQL database; hence, there are no features specific to the domain of Event Sourcing, such as event ordering or projections. As a developer, you will need to decide how to implement these.
In true AWS fashion, the documentation for DynamoDB is not top notch. While the learning curve is generally very soft and it's not hard to learn, you need to have at least some experience with cloud and database management to be able to start using DynamoDB and understand the documentation.