Changes include:
 - A new configuration layout for supporting multiple sources, sinks and mappings.
 - No more enqueuing delay. Queues tend to either be full or empty, and are really only effective at smoothing jitter. Spinning to delay doesn't really help, so we just drop the events immediately instead of blocking.
 - No more session-binning strategy for HDFS. In practice we didn't use this, and it's complicated to do properly when multiple sources (with their own idea of sessions) might be feeding into a sink.

For now, the code only supports the existing single browser-based source and HDFS/Kafka sinks. Configuration validation is also not yet implemented.

…mapping exceptions and turn them into useful error messages and leave configuratio invalid.

# Conflicts:
#	build.gradle

…operty is not optional. This is to make the constructor args match the reality of the config params, such that we can later use JavaDoc for the documentation. Default values are set using the defaultValue annotation, which is however not interpreted by the deserializer. We will later fix this using a custom deserializer in Jackson.

…or into dev/many-to-many

# Conflicts:
#	src/main/java/io/divolte/server/kafka/KafkaFlusher.java
#	src/main/java/io/divolte/server/kafka/KafkaFlushingPool.java
#	src/main/resources/reference.conf

All users of this class did this anyway.

 - Eliminate a bunch of empty configuration files.
 - Change test layout to let tests run against the default configuration with the test server, or a base convenience configuration.

The browser source requires this or the tracking script won't correctly locate the event end-point.

No trailing period (.) required.

It's invoked by the constructor, and constructors shouldn't normally invoke instance methods because they might not expect the instance to be partially constructed.

…ing into.

Changes include:
 - Sink configurations now need to supply a factory for creating the sink.
 - The optional GeoIP service is now passed around as such. It's a bit weird why it was nullable in some places before.
 - There's now a central point for loading schemas. This is needed for schema inference with sinks and ensures schemas are only loaded once.
 - Some utilities for collecting streams into Guava's immutable collections.
 - Processing pools are now shutdown by the central server, instead of recursively via the mapper.

The associated tests have also been updated; the fixture for the previous test was not correct.

The main code already depends on these artefacts.

…e version.

… shared sinks.

…class.

This allows source initialisation to take place in parallel, a win because it tends to be fairly slow.

Somewhat embarrassingly, the mapping hadn't been converted to groovy yet. (The mapping hasn't been tested.)

… configuration for consistency.

Retrieving source configurations is now the same as for sinks. In addition, we now universally throw IllegalArgumentException if the the wrong name and/or type is used. (This is a programming mistake, not a user error, as the javadoc now indicates.)

Also substitute some method handles.

# Conflicts:
#	build.gradle

…initialisation.

…est.

…ble, and if it's not that's exceptional.

…ayout.

…tion format.

… version of Kafka that we're using.

…appings in Divolte.

… configured.

…ript is specified.

Changes include Sphinx markup, and updates to reflect that we now support more than a single mapping.

Plus a few typo corrections.