Kafka sources that don't read from any partition never emit a watermark,
thereby blocking the progress of event-time in downstream operations.
This changes the Kafka Source to emit a Long.MAX_VALUE watermark if it
knows that it will never receive data.

This also changes the Timestamp Extraction operator to reacto to a
Long.MAX_VALUE watermark by itself emitting a Long.MAX_VALUE watermark.

Before, would throw a ClassCastException when emitting watermarks with
timestamp/watermark multiplexing disabled.

This closes #1416

This closes #1441.

- This solves errors with reflectasm using Scala 2.11 and Java 8

This closes #1445.

This closes #1443.

Before, streaming mode (either batch or streaming) would specify how
memory is allocated on task managers.

This introduces a new configuration value taskmanager.memory.allocation
that can take values "lazy" or "eager". This controls how memory is
allocated.

This closes #1437

This closes #1440

This closes #1435

This closes #1436

This closes #1292

[FLINK-2488] [FLINK-2524] [FLINK-3124] Expose Attempt Number in RuntimeContext and add TaskInfo to hold all task related parameters.

This closes #1386

This closes #1366

This closes #1430

It seems AbstractJobVertex class no longer exists but the doc's link still points to it.

This closes #1438

Added Time.of(Time), because there is no TumblingTimeWindows.of(int, TimeUnit).

New users sometimes struggle with the imports (especially for Scala API).

This closes #1409

This closes #1431.

This closes #1383

This also adds a test for asynchronous state handles.

Before, it would not allow unioning with predecessors (also transitive)
and streams of differing parallelism.

- refactor to use Storm's topology builder
- remove FlinkTopologyBuilder
- instantiate context-based StreamExecutionEnvironment (local or remote)
- remove some of the Flink and Storm behavior replicating classes
- modify FlinkTopology to parse Storm topology directly
- replace StormTestBase with StreamingTestBase
- add print example
- FlinkTopologyBuilder changes (check if all inputs are available before processing)
- correct package typo
- two input support
- add join example
- update docs

This closes #1398.

The RabbitMQ source may operate in three different modes:

1) Exactly-once (when checkpointed) with RabbitMQ transactions and messages with
   unique correlation IDs.

2) At-least-once (when checkpointed) with RabbitMQ transactions but no
   deduplication mechanism (correlation id is not set).

3) No strong delivery guarantees (without checkpointing) with RabbitMQ
   auto-commit mode.

- add base class which can handle both session ids and unique ids
  - session ids are used for acknowledgment
  - unique ids are used for de-duplication

- add unit test

- add documentation

This closes #1243.

This closes #1424.