Run the symm staging fill on byte views (no fp8 index_copy kernel)

First e2e launch crashed every CP rank at the symm token fill:
index_copy_cuda is not implemented for Float8_e4m3fn — the production
KV pool dtype, which the 8-rank test missed by building its pools as
uint8.  The fill is a whole-token-row copy, so it is dtype-agnostic:
both the staging span and the current rows now go through uint8 views.

The 8-rank test now builds the KV pools and current rows as
float8_e4m3fn (payloads constructed as bytes, compared as bytes) so
the production dtype is what every phase exercises.

Validated on g0033: all four 8-rank byte-exactness phases under fp8.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

python/sglang/srt/layers/attention/nsa/cp_shared_kv_prefetch.py

@@ -1001,14 +1001,18 @@ class CpSharedKVMlaPrefetcher:
             )
             num_rows = int(state.staging_current_rows.numel())
             if num_rows > 0:
-                staging_rows = staging_span.view(kv_cache.dtype).reshape(
+                # Byte view on both sides: index_copy_ has no fp8 CUDA
+                # kernel, and the copy is dtype-agnostic (whole token rows).
+                staging_rows = staging_span.view(
                     (state.num_current_pages + 1) * self.page_size,
-                    *kv_cache.shape[1:],
+                    state.page_nbytes // self.page_size,
                 )
                 staging_rows.index_copy_(
                     0,
                     state.staging_current_rows,
-                    current_kv_cache[:num_rows],
+                    current_kv_cache[:num_rows]
+                    .reshape(num_rows, -1)
+                    .view(torch.uint8),
                 )
             cp_symm_barrier(
                 staging.flag_ptrs, self_rank=int(self.layout.cp_rank)

python/sglang/srt/layers/attention/nsa/cp_shared_kv_runtime.py

@@ -4876,14 +4876,18 @@ def _compose_token_kv_partial_current_v2(
         )
         num_rows = int(fill_state.staging_current_rows.numel())
         if num_rows > 0:
-            staging_rows = staging_span.view(kv_cache.dtype).reshape(
+            # Byte view on both sides: index_copy_ has no fp8 CUDA kernel,
+            # and the copy is dtype-agnostic anyway (whole token rows).
+            staging_rows = staging_span.view(
                 (int(plan.num_current_pages) + 1) * page_size,
-                *kv_cache.shape[1:],
+                kv_page_nbytes // page_size,
             )
             staging_rows.index_copy_(
                 0,
                 fill_state.staging_current_rows,
-                current_kv_cache[:num_rows],
+                current_kv_cache[:num_rows]
+                .reshape(num_rows, -1)
+                .view(torch.uint8),
             )
         _symm_barrier_and_gather_all(
             kernels=kernels,