OpenSearch source should write to buffer directly (no BufferAccumulator)

The opensearch source is using the BufferAccumulator to write to the buffer. But, this is an unnecessary structure because the source already queries the OpenSearch cluster/domain/collection in batches. Each batch should be a single Buffer::writeAll.

This will help the metrics align between batches and buffer writes. And batches will be all-or-nothing similar.

The BufferAccumulator was made for the S3 source. This had a stream of data coming in and the number of events is unknown to the sink until after it is parsed. The BufferAccumulator allows for batches writing to the buffer. It is only appropriate for situations where there is no natural batch available. For example, the http source has batches pushed from clients and the opensearch source searches in batches.

Start by locating the OpenSearch source and its use of BufferAccumulator, then inspect the surrounding batch-query and buffer-writing flow. Replace the accumulator path with one Buffer::writeAll per OpenSearch batch, and verify that batch metrics align with buffer writes and writes remain all-or-nothing using the relevant existing tests.