fix: partition update

[ccm32004]

updates numaflow-java, due to changes made in numaflow

Found issues outlined in the current getPartition implementation, thus some changes were made in how the getActivePartition method is implemented.

Why watermarks weren't updating properly, causing the window to not close properly

• Old getActivePartitions() (or deprecated method getPartitions) returned all Pulsar topic partitions ([0..4]), making Numaflow spin up 5 watermark processors per replica.
• Under Pulsar's Shared subscription, messages from any partition can land on any consumer replica. There's no stable replica -> partition ownership, so multiple replicas end up claiming the same watermark processor IDs with inconsistent progress.
• Low TPS (1 msg / 5s): some partitions never receive data on a given replica, so those processors stay at -1 and the min is pinned low
• High TPS (10k rpu): all partitions receive data, but it's spread across replicas dynamically, so each replica only advances the processors whose messages it happened to see — gaps still stall the min.

Why this fix works

• getActivePartitions() now returns defaultPartitions() — a singleton with this replica's NUMAFLOW_REPLICA index, so the watermark partition is the replica, not a Pulsar topic partition.
• Each replica owns exactly one watermark processor that it fully controls, so there is no ambiguity, no shared/gapped processors, etc
  Matches Numaflow's upstream Rust Pulsar source, which has always modeled things this way.

fixes #87

[github-actions]

Continuous benchmark

Config: batch size = 500

Compared to the latest result on gh-pages (last main run that updated charts). Regression when a metric degrades by more than 2× vs baseline.

Metric	This run	Baseline (main)	Change	Status
Consumer Throughput	34566.67 msgs/sec	42005.56 msgs/sec	-17.7%	OK
Processing Latency (per batch)	14.48 ms	11.91 ms	+21.6%	OK
Read Latency (per batch)	7.39 ms	5.72 ms	+29.2%	OK
Ack Latency (per batch)	0.50 ms	0.48 ms	+4.2%	OK
Consumer CPU	1332 millicores	1443 millicores	-7.7%	✅ Improved
Consumer Memory	372 MiB	356 MiB	+4.5%	OK

Result: No regression beyond the threshold vs main baseline.

Workflow run
· GitHub Pages charts

[github-actions]

Code Coverage

Percentage of source code lines executed by unit tests.

Overall Project	82.25%	✅
Files changed	100%	✅

File	Coverage
PulsarSource.java	76.92%	✅