Removed flushing of NTA subtrees and dependency transferral.

Flushing NTA changes on dependency change is unnecessary work.

Removed commented-out debug code to clean up the template and generated code.

An NTA is never primitive or of type String.

Removed commented-out code to clean up aspect.

Added new attribute AttrDecl.isHOA() to tests if an attribute declared
as an NTA either in the aspect file or in the abstract grammar.

Removed commented-out debug code to clean up the template and generated code.

Extract method to prepare for larger refactoring.

Remove redundant statements by moving out of cases in if-statement.

Release management is being moved to a separate repository.

Improve dependency tracking and flushing for incremental collection attributes
by treating the survey phase as an attribute with its own dependency graph
node. This makes all collection attribute instances depend on the survey phase
and any change which flushes the survey results leads to flushing all
collection attribute instances as well.

fixes #302 (bitbucket)
fixes #303 (bitbucket)

Use generic type argument for ArrayList to simplify iteration.

Allow implicit root type for collection attributes when the grammar has
multiple root types.

If there are multiple roots in the grammar, the collection root type is set to
ASTNode.

Factored out root finding in collection attribute templates.

fixes #294 (bitbucket)

When there was not an equation for some NTA on an AST class, but the NTA hand
an equation in some superclass, the NTA was incorrectly accessed through the
child vector instead of having a special accessor through getChildNoTransform().
This error was just caused by not looking for NTA equations in superclasses
in ASTDecl.hasSynEq().

fixes #299 (bitbucket)

Avoid generating a child access for NTAs which don't have an equation on the
current class.

fixes #298 (bitbucket)

This escapes unintentional Unicode escape sequences in filenames (on Windows,
for directories starting with 'u').

fixes #297 (bitbucket)

Improved output of --statistics option:

* List all classes in CSV (added non-grammar classes).
* Added is_ast column (TRUE/FALSE) for discriminating AST classes from non-grammar classes.
* Added columns for refined syns, inhs, and ITDs (refinedsyn, refinedinh, refineditd).

Given that collection attributes are computed in batch, the first collection attribute
instance computed for a group will collect all the dependencies for the group of
collection attributes. Hence, at time of flush, the whole group of collection attributes
must be flushed and not only the notified initial collection attribute instance driving
the evaluation for the group.

In the case where a collection attribute is not defined on the root used
to collect contributors, then the root needs to be located in order to
reset the list of contributors on change propagation.

The relative node ID should be printed when possible and not
the node ID.

For cases when a list is created via the List(T... ) constructor. This constructor
creates a child array and then calls addChild (which calls setChild). In setChild,
the creation of the child handler is controlled by whether the child array is null.
Creating the child handler in this constructor ensures that it is no null.

Updates incremental code generation to comply with refactoring of generated code for parameterized NTAs.

Parameterized NTAs used to have a List node as parent but this has been changed to a plain ASTNode
referred to as a proxy node. The incremental implementation generated additional code for Lists to
distinguish internal NTA lists from normal lists and made use of iteration over list children.

The incremental construction state should not be entered for these edit operations
as such edits should trigger change propagation. The construction state should only
be entered when constructing the initial AST from constructors or when modifying the
AST during CTNA evaluation or similar.

This method may be called recursively for rewritten nodes.

This check is probably connected to legacy rewrites. With rewrites implemented
as circular NTAs, a change notification to the node with the rewrittenNode
attribute should lead to a flush of the rewrittenNode attribute regardless of
whether the rewritten node is a region root or not.

Entering the construction state avoids change propagation for edits
occuring during evaluation of circular NTAs.

Handlers are added as listeners when regions are crossed, that is, when parents
and children at the region edge are accessed. After such a crossing values in the
entered region may be used freely, hence the region accessing the neighbouring
region is added as a listener and is also marked as computed.

The construction state should only be entered in the constructor, other
AST modification methods should trigger notification.

- Adds enter/exit construction for List constructor.
- Adds handler check to setChild for region to prevent duplicate notificaiton on the same handler.
- Adds check of computed OR created to handler when reacting to dependency change.

- Adds handle state updates: empty and computed.
- Adds state to handler string.
- Simplifies generated code by removing on intermediate method call.