fix(vpto): use typed vstus LLVM callees (#1124)

[Zhendong404]

Summary

• Make vstus LLVM callees type-specific in both VPTO emitters.
• Add mixed-type vstus regression coverage for default and CANN 9.0 lowering.
• Update the low-precision memory regression expectation to the typed callee.

Validation

• ninja -C builds/issue-1124 check-dsl — 25/25 passed.
• Focused vstus FileCheck tests — passed with the build-tree compiler.
• Full check-pto — 1603 passed, 1 unsupported, 4 pre-existing tadd FileCheck failures caused by SSA constant-name suffixes.

Closes #1124

[reedhecre]

A5 板测失败

• 触发方式：merged
• 源码提交：161a71657e14
• 结果汇总：OK 0 / FAIL 21 / SKIP 0
• 日志：/root/ptoas-board-monitor-a5/logs/20260806_191905_merged_pr1136.log
• 失败阶段：board-validation-qwen / exit=1

失败用例

• rope_kv_cache (run, exit=1)
• rmsnorm (run, exit=1)
• qwen3_decode_incore_7 (run, exit=1)
• qwen3_decode_incore_6 (run, exit=1)
• qwen3_decode_incore_5 (run, exit=1)
• qwen3_decode_incore_4 (run, exit=1)
• qwen3_decode_incore_2 (run, exit=1)
• qwen3_decode_incore_1 (run, exit=1)
• qwen3_decode_incore_12 (run, exit=1)
• qwen3_decode_incore_11 (run, exit=1)
• qwen3_decode_incore_10 (run, exit=1)
• post_rmsnorm (run, exit=1)
• out_proj_residual (run, exit=1)
• down_proj_residual (run, exit=1)
• sparse_attn_test_incore_7 (run, exit=1)
• decode_swa_test_incore_40 (run, exit=1)
• decode_hca_test_incore_54 (run, exit=1)
• decode_csa_test_incore_81 (run, exit=1)
• attention_swa_test_incore_40 (run, exit=1)
• attention_hca_test_incore_54 (run, exit=1)
• attention_csa_test_refresh_incore_81 (run, exit=1)

[reedhecre]

A3 板测完成（有跳过）

• 触发方式：merged
• 源码提交：161a71657e14
• 结果汇总：OK 220 / FAIL 0 / SKIP 3
• 日志：/home/zhongxuan/ptoas-board-monitor/runtime/logs/20260806_054938_merged_pr1136.log
• 结果 TSV：/home/zhongxuan/ptoas-board-monitor/runtime/logs/20260806_054938_merged_pr1136.tsv