[RFC] support accumulator in cpgpass without API changes

[max-leuthaeuser]

Rewrote ForkJoinParallelCpgPassWithAccumulator to use the stream.collect(Supplier, BiConsumer, BiConsumer) API instead of ThreadLocal + ConcurrentLinkedQueue.

1. Combined container (BuilderWithAccumulator): Pairs a DiffGraphBuilder with an accumulator R per fork. The Supplier creates one per thread, so both are naturally thread-local without explicit ThreadLocal management.
2. Three stream.collect callbacks mirror ForkJoinParallelCpgPass:
   - Supplier: creates a fresh BuilderWithAccumulator(newDiffGraphBuilder, newAccumulator())
   - Accumulator BiConsumer: calls runOnPartWithAccumulator(bwa.builder, bwa.acc, part)
   - Combiner BiConsumer: absorbs the right builder into the left AND merges accumulators via mergeAccumulators
3. Error handling preserved: Uses _accResult/_hasResult to store the merged accumulator. finish() calls onAccumulatorComplete with the merged result on success, or a fresh newAccumulator() on failure.
4. Extends CpgPassBase directly instead of ForkJoinParallelCpgPass to avoid an unused/dead runOnPart method.

Hence: no API changes to the original ForkJoinParallelCpgPass, no null/Null, still using the pretty stream.collect(Supplier, BiConsumer, BiConsumer) API.
Downside: some code duplication. That's unavoidable if you don't want to change the existing API.

[max-leuthaeuser]

I will create a new PR merging all the stuff we discussed.