docs: fix stale embedding probe batch sizes and CLI profile activation

docs/configuration.md

@@ -84,7 +84,7 @@ Environment precedence is universal across Make targets and ingestion scripts:
 
 ### Embedding preflight checks
 
-Before indexing, ingestion scripts verify the model alias through `/v1/models`, then issue `X-Tier: batch` embedding probes with batches of 1 and 4. Each response must preserve input count/order and contain exactly 2,560 numeric values per vector. Any HTTP, model, shape, null-value, or dimension failure stops ingestion.
+Before indexing, ingestion scripts verify the model alias through `/v1/models`, then issue `X-Tier: batch` embedding probes with batches of 1 and 8. Each response must preserve input count/order and contain exactly 2,560 numeric values per vector. Any HTTP, model, shape, null-value, or dimension failure stops ingestion.
 
 ## Qdrant
 

docs/github-repository-ingestion.md

@@ -97,7 +97,7 @@ An embedding or upsert failure leaves the prior complete page and its marker int
 ## Failure diagnostics and retry behavior
 
 - GitHub ingestion fails fast when embedding or vector writes fail.
-- Preflight validates the gateway model alias and `X-Tier: batch` embedding batches of 1 and 4 before ingestion starts.
+- Preflight validates the gateway model alias and `X-Tier: batch` embedding batches of 1 and 8 before ingestion starts.
 - When null/invalid vectors are detected, diagnostics now state likely causes explicitly:
   - wrong endpoint (must resolve to `/v1/embeddings`)
   - non-embedding model

docs/pipeline-commands.md

@@ -267,7 +267,7 @@ set +a
 
 # With doc set filtering
 DOCS_DIR=/absolute/path/to/docs DOCS_SETS=spring-framework-reference \
-  java -Dspring.profiles.active=cli -jar "$app_jar" \
+  java -Dspring.profiles.active="cli,$SPRING_PROFILE" -jar "$app_jar" \
   --spring.main.web-application-type=none --server.port=0
 ```