Switch to using objects instead of classes for usercode

Whole stack is changed so that input/output formats and PACT contracts,
respectively their builders, accept objects in addition to classes for
user code stubs.

So now one can do:
MapContract.builder(new CountWords())
in addition to:
MapContract.builder(CountWords.class)

Objects passed in this way need to be serializable. The advantage is
that one can now pass arguments in the constructor of the stub, store
them in the stub, and they will be available when the stub is run. This
is less cumbersome than using the Configuration to pass parameters.

This change was originally  necessary to allow the scala frontend to generate
usercode objects to pass to PACT contracts.

This is needed for the magic that happens in the scala frontend.

This is needed because the scala frontend will create a custom
PartialSolutionHolder to allow it's magic to happen.

Pact Compiler fix+testcase for SinkJoinerPlanNode

With help from @uce

Support for different Hadoop versions (yarn!) using activation by properties

Remove TeraSort from examples

Remove unused exclude-statements in assembly-descriptor (exclude elements can only define artifacts, not paths)

1) Changed profile activation to properties (like hbase) 2) moved all properties to top of pom 3) moved module list to top of pom

Added ALL_GROUP strategy to PlanJSONDumpGenerator.

Added ALL_GROUP strategy to PlanJSONDumpGenerator. Before this fix the pact web could not show the plan.

Fixed nephele-config.sh substr command calls on OS X (closes #88)

Fixed bug. stub.open() was not called for COMBININGSORT local strategy.

Fixed commons-io missing in lib directory + fixed wrong groupId of commons-io

1) Fixed problem where commons-io jar file was missing in the lib directory 2) Fixed wrong maven group definition of commons-io (changed from org.apache.commons to commons-io) and updated to a more recent version (same as hadoop)

updated contributers

added comment for eclipse users

Remove pact-examples dependency in pact-clients.

Made the hadoop_v1 maven profile active by default (hadoop version 1.2.1)

Hopefully fixed a bug in the pact compiler for branching plans:
	- added a test case provided by Anja Kunkel from HU Berlin
	- my understanding of the bug is the following: During a check if additional pipeline breakers are necessary,
	the compiler checks for the LocalStrategy of a SinkJoinerPlan. The SinkJoinerPlan is a "fake" plan element to group multiple output sinks into one sink (the compiler assumes one root node, this is why a "fake" plan node is required).
	The local strategy of SinkJoinerPlan is initialized as NONE, which means that we just pass through the elements (which makes sence since we are talking about a sink) The initialization of a NONE-local strategy assumes only one input, not two.
	I replaced the LocalStrategy with BINARY_NO_OP, which is a NONE for two input.
	The compiler is now able to do its pipline breaker checks.

Update Readme.md with project overview and a little tutorial.

Fix from Alexander, so that pact-hbase is only included if yarn is used (because it is not compatible with hadoop v1)