It would be good for all JavaCC languages to share most of their implementation, so that changes like #2166, #1446 or #2195 are propagated to all modules. To be able to change things across all languages with minimal trouble, merging their Ant build files is essential.

This PR creates a new ant build file and makes pmd-java & pmd-modelica use it as a PoC. Other modules will be ported one by one.

Changes that using the new build does to a module:

• Use the ParseException of pmd-core instead of duplicating a class. Error messages of that one are slightly better.
• No JJT${langname}ParserState is generated. The one in pmd-core, named JjtreeBuilder, is used instead. This is an improved version that features the changes needed by java-grammar to support annotations, and the changes of [modelica] Normalize invalid node ranges #2195 about implicit tokens.
• The constant interface for token ids is turned into a final utility class. Eg JavaParserConstants become JavaTokenKinds. Lexical state IDs are not made public.
• The constant interface for node ids is made package-private. Ids have a single purpose (implementing getXpathNodeName) and this is better-off hidden.
• Use JavaccToken instead of generating a new Token class
• Tokens are created by a TokenDocument. This is useful to avoid writing too much logic inline in lexical actions. Eg JavaTokenFactory is now JavaTokenDocument. Later a token document could support edition methods, to implement autofixes
• The generated visitor interface has default methods (#1444).
• The build file deletes the node files that were found in the main source tree. This avoids hardcoding their names as is currently done in pmd-modelica or pmd-plsql

Conventions that the build file relies on:

• The grammar file name is the same as the language name. Eg Java.jjt, but PldocAST.jjt will be renamed PLSQL.jjt
• The generated parser is named ${lang-name}ParserImpl. This is something that has bugged me for a while: many modules have several public parser classes with the same name, eg JavaParser or PLSQLParser. Finding the file in an IDE is a pain and we must use FQCNs. I think the only published API should be the implementation of our Parser interface. So JavaLanguageParser is renamed JavaParser, moved into the ast package, and the generated parser is named JavaParserImpl and is package-private.

Nodes don't use the constructors taking the parser anymore so I removed them. Since parsers change names, it would have made a big diff to update the constructors for no benefit. The parser was previously only used to fetch the current token in jjtClose. This is a "bug" of JJTree: it sets the last token after jjtClose. The new JjtreeBuilder sets it correctly before, so in jjtClose the node has all the information available.

All of this is only possible if we somewhat stop relying on the JavaCC generated helper classes (eg char streams), and so some are moved to the main source tree of pmd-core. We share the implementation, and can modify them and document them independently of JavaCC.

This will be a series of PRs so I'll open an issue to track them. We should also take care of describing these conventions on the wiki & website