fix(specs): clarify decompounding limitations (#3227)

specs/common/schemas/IndexSettings.yml

@@ -144,6 +144,8 @@ baseIndexSettings:
         You can specify different lists for different languages.
         Decompounding is supported for these languages:
         Dutch (`nl`), German (`de`), Finnish (`fi`), Danish (`da`), Swedish (`sv`), and Norwegian (`no`).
+        Decompounding doesn't work for words with [non-spacing mark Unicode characters](https://www.charactercodes.net/category/non-spacing_mark).
+        For example, `Gartenstühle` won't be decompounded if the `ü` consists of `u` (U+0075) and `◌̈` (U+0308).
       default: {}
       x-categories:
         - Languages
@@ -527,10 +529,12 @@ indexSettingsAsSearchParams:
     decompoundQuery:
       type: boolean
       description: |
-        Whether to split compound words into their building blocks.
+        Whether to split compound words in the query into their building blocks.
 
         For more information, see [Word segmentation](https://www.algolia.com/doc/guides/managing-results/optimize-search-results/handling-natural-languages-nlp/in-depth/language-specific-configurations/#splitting-compound-words).
         Word segmentation is supported for these languages: German, Dutch, Finnish, Swedish, and Norwegian.
+        Decompounding doesn't work for words with [non-spacing mark Unicode characters](https://www.charactercodes.net/category/non-spacing_mark).
+        For example, `Gartenstühle` won't be decompounded if the `ü` consists of `u` (U+0075) and `◌̈` (U+0308).
       default: true
       x-categories:
         - Languages