kie-issues#2229: process codegen method split - #6879

…s (WIP)

Transplants every method ProcessVisitor generates onto the process wrapper
class instead of just the first (ProcessGenerator.addProcessMethods),
threads the containing class into ProcessVisitor.visitProcess(), and splits
process()'s single monolithic method into initVariables()/initMetadata()/
initNodes()/initConnections() so process() itself stays O(1) regardless of
process size.

Verified against the real repro-fails.bpmn / repro-control-passes.bpmn
reproducer (apache/incubator-kie-issues#2229): Tier 1 alone is NOT
sufficient - initNodes() still overflows the JVM 64KB per-method limit for
both files once real BPMN diagram/metadata weight is included (not just the
programmatically-built RuleFlowProcessFactory case, which is why
ProcessCodeTooLargeReproducerTest currently fails - that is expected and
intentional, it is the evidence that Tier 2 (chunking within initNodes(),
statement-count-based per the metric comparison against apache#4176's rejected
node-count approach) is required, not optional).