[Bugfix] Fix masked_fill_ function typo (#769)

### What this PR does / why we need it?
Fix function name typo, make `mask_fill_` to `masked_fill_`

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
CI passed

Signed-off-by: ApsarasX <apsarax@outlook.com>

Author: ApsarasX

vllm_ascend/attention/attention.py

@@ -135,7 +135,7 @@ def get_splitfuse_attn_mask(
                       context_len:] = self.splitfuse_mask_value
             right_tensor = attn_mask[current_row:current_row + q_len,
                                      context_len:seq_len]
-            right_tensor.mask_fill_(
+            right_tensor.masked_fill_(
                 right_tensor.tril() == self.splitfuse_mask_value, 0)
             current_row += q_len