docs(example): clarify nonpad_kv_seqlen is the after-scatter valid count

Update the module-docstring bullet to match the fix's after-scatter semantics:
nonpad_kv_seqlen is the number of valid KV entries after the current chunk is
scattered (write_indices + cur_seq_len), not a generic running count.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: titaiwang <titaiwang@microsoft.com>

Author: titaiwangms

examples/static_cache_generation.py

@@ -13,7 +13,8 @@
 - No ``attention_mask`` input — causal masking is handled internally.
 - 3-D cache shape ``[batch, max_seq_len, kv_hidden]`` (not 4-D).
 - ``write_indices`` tracks where to write the next token's KV entry.
-- ``nonpad_kv_seqlen`` tracks how many valid KV entries exist.
+- ``nonpad_kv_seqlen`` is the number of valid KV entries after the current
+  chunk is scattered into the cache (``write_indices + cur_seq_len``).
 - Outputs are ``updated_key_cache.{i}`` / ``updated_value_cache.{i}``.
 
 Usage::