Fix RDS joins: set workers=1 on main pipeline to prevent event reordering (opensearch-project#6784)

The joins template inherited the user-configured workers count (default 2)
for the main pipeline. With workers > 1, multiple threads write to the S3
sink concurrently. The S3 sink's ReentrantLock serializes writes but does
not guarantee ordering — thread 2 can write item2 before thread 1 writes
item1 for the same parent document.

When the S3 sub-pipeline reads these out-of-order events and sends them
to OpenSearch, the per-table version check in the Painless script rejects
the lower-versioned item (noop), causing data loss for 1:N child records.

Setting workers=1 ensures events are written to S3 in binlog order.
This has no throughput impact since the S3 sink's ReentrantLock already
serializes writes to a single thread at a time.

Tested with 200 threads, 5M orders: 0 failures with workers=1 vs
~0.09% failure rate with workers=2.

Signed-off-by: Dinu John <86094133+dinujoh@users.noreply.github.com>

Author: dinujoh

data-prepper-plugins/rds-source/src/main/resources/org/opensearch/dataprepper/transforms/templates/rds-joins-template.yaml

@@ -8,7 +8,7 @@
 #
 
 "<<pipeline-name>>":
-  workers: "<<$.<<pipeline-name>>.workers>>"
+  workers: 1
   delay: "<<$.<<pipeline-name>>.delay>>"
   buffer: "<<$.<<pipeline-name>>.buffer>>"
   source: