[PIR]Fix optional output error when input is none

[0x45f]

PR types

Bug fixes

PR changes

Others

Description

修复optional 输出的bug，以masked_multihead_attention_为例

- op : masked_multihead_attention_
  args : (Tensor x, Tensor cache_kv, Tensor bias, Tensor src_mask, Tensor cum_offsets, Tensor sequence_lengths, Tensor rotary_tensor, Tensor beam_cache_offset, Tensor qkv_out_scale, Tensor out_shift, Tensor out_smooth, int seq_len, int rotary_emb_dims, bool use_neox_rotary_style=false, str compute_dtype = "default", float out_scale=-1, int quant_round_type=1, float quant_max_bound=127.0, float quant_min_bound=-127.0)
  output : Tensor(out), Tensor(cache_kv_out), Tensor(beam_cache_offset_out)
  infer_meta :
    func : MaskedMultiheadAttentionInferMeta
  kernel :
    func : masked_multihead_attention
    data_type : x
  optional : bias, src_mask, cum_offsets, sequence_lengths, rotary_tensor, beam_cache_offset, qkv_out_scale, out_shift, out_smooth
  inplace : (cache_kv -> cache_kv_out), (beam_cache_offset -> beam_cache_offset_out)

第三个输出beam_cache_offset_out是optional的，它会和输入beam_cache_offset做inplace，输入beam_cache_offset是optional 的，当输入beam_cache_offset为None时，执行器会报错。在此PR中对这种情况做了处理，当beam_cache_offset为None时，将beam_cache_offset_out的type设置为null type。会在API内生成类似下面的代码

  if (!beam_cache_offset) {
    masked_multihead_attention__op.result(2).set_type(pir::Type());
  }

Pcard-67164