Add option for splitting generated Java parser into 3 files

[ericvergnaud]

The current generator for Java embeds, within the Parser class, the serialized ATN and ParserRuleContext subclasses (as static child classes).

With large or complex grammars, this can lead to very large parser files that exceed 65k lines.
This is a problem because the JVM does not accept line numbers > 65535, thus making it impossible to debug these parsers.

This PR fixes the problem by adding a -split-parser option, that instructs the Tool to generate the serialized ATN and the ParserRuleContext subclasses in dedicated files, thus dramatically reducing the number of lines of the generated Parser.

For the maven plugin, this option can be set using the <arguments>-split-parser</arguments> configuration entry.

There is no change if this option is not used.

[parrt]

Nothing seems crazy, but of course it's a widespread change. I'm just going to have to trust you. :) should we update the documentation to include the new options for Java target?

[ericvergnaud]

Thanks