[Docs] Follow the upstream op list: drop the packed GQA prefill op, add FusedTopKOp

docs/api/attention.md

@@ -33,12 +33,6 @@ what runs when you call `op(...)`.
       heading_level: 3
       members: ["__init__", "forward"]
 
-::: tileops.attention.GroupedQueryAttentionPrefillFwdOp
-    options:
-      show_root_heading: true
-      heading_level: 3
-      members: ["__init__", "forward"]
-
 ::: tileops.attention.GroupedQueryAttentionPrefillVarlenFwdOp
     options:
       show_root_heading: true

docs/api/moe.md

@@ -6,10 +6,10 @@ Both are documented under each op — `__init__` and `forward`, where `forward`
 what runs when you call `op(...)`.
 
 A routed mixture-of-experts layer is available two ways here. `FusedMoeFwdOp` runs
-the whole FFN. The rest are its stages, callable on their own: the ops that move
-tokens into an expert-contiguous layout and back, and the expert GEMMs that run on
-it. The GEMMs come in a padded form and a tight one, and the routing has to produce
-the layout the GEMM expects.
+the whole FFN. The rest are its stages, callable on their own: the op that picks
+each token's experts, the ops that move tokens into an expert-contiguous layout and
+back, and the expert GEMMs that run on it. The GEMMs come in a padded form and a
+tight one, and the routing has to produce the layout the GEMM expects.
 
 ## Fused forward
 
@@ -21,6 +21,12 @@ the layout the GEMM expects.
 
 ## Routing and layout
 
+::: tileops.moe.FusedTopKOp
+    options:
+      show_root_heading: true
+      heading_level: 3
+      members: ["__init__", "forward"]
+
 ::: tileops.moe.MoePrePermuteFwdOp
     options:
       show_root_heading: true