Fixes https://github.com/spring-projects/spring-integration-aws/issues/90
* Add an internal `ShardConsumerManager` which is responsible to initiate
a locking for the shard key in the provided consumer group and populating
a `ShardConsumer` if `tryLock()` is successful or no `LockRegistry` at all
* Additional logic is added to always iterate over candidate shards if
`tryLock()` on the matter is not successful.
This way the current `KinesisMessageDrivenChannelAdapter` picks up
those shards which have been locked by the consumer which has just
left a cluster and unlocked distributed locks
* Now all the shards are considered as candidates independently of the
`streams` or `shardOffsets` configuration
* Improve Kinesis tests performance
0e02b0d470