Expands bbq_hnsw known issue (elastic#132539)

* Expands bbq_hnsw known issue.

* Refines text.

* Addresses feedback.

* Update docs/release-notes/known-issues.md

* Addresses feedback.

Author: szabosteve

docs/release-notes/known-issues.md

@@ -5,13 +5,19 @@ mapped_pages:
 ---
 
 # Elasticsearch known issues [elasticsearch-known-issues]
+
 Known issues are significant defects or limitations that may impact your implementation. These issues are actively being worked on and will be addressed in a future release. Review the Elasticsearch known issues to help you make informed decisions, such as upgrading to a new version.
 
 ## 9.1.0 [elasticsearch-9.1.0-known-issues]
-* An error in the configuration of vector indices with type `bbq_hnsw` may lead to significant search performance degradation on 9.1.0. To mitigate this, set the `-Dvector.rescoring.directio=false` JVM option on all search nodes, then restart the nodes.
-  This option can be removed in 9.1.1.
+
+* The `-Dvector.rescoring.directio` JVM option is enabled (set to `true`) by default. When used with `bbq_hnsw` type vector indices, this can cause significant search performance degradation; particularly when enough memory is available to hold all vector data. In some cases, kNN search latency can increase by as much as 10x. To mitigate this, set the JVM option `-Dvector.rescoring.directio=false` on all search nodes and restart them. This option can be removed in 9.1.1.
+
+**How do I know if my index vector type is `bbq_hnsw`?**
+
+* Prior to 9.1, the vector type had to be explicitly set to `bbq_hnsw`. Starting with 9.1, `bbq_hnsw` is the default vector type for dense vectors with more than 384 dimensions in new indices, unless another type is specified.
 
 ## 9.0.3 [elasticsearch-9.0.3-known-issues]
+
 * A bug in the merge scheduler in Elasticsearch 9.0.3 may prevent shards from closing when there isn’t enough disk space to complete a merge. As a result, operations such as closing or relocating an index may hang until sufficient disk space becomes available.
 To mitigate this issue, the disk space checker is disabled by default in 9.0.3 by setting `indices.merge.disk.check_interval` to `0` seconds. Manually enabling this setting is not recommended.