[WIP] multiheadattention optional q k v out affine

[nihui]

• optional q k v out affine aka. sdpa
• pnnx ncnn sdpa
• x86 sdpa
• arm sdpa
• vulkan sdpa
• test sdpa
• test sdpa int8
• mqa and gqa
• pnnx ncnn mqa and gpa fusion
• x86 mqa and gqa
• arm mqa and gqa
• vulkan mqa and gqa
• test mqa and gqa
• test position embedding
• test latent attention

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 60.48387% with 49 lines in your changes missing coverage. Please review.

Project coverage is 95.36%. Comparing base (2389090) to head (94039d8).
Report is 3 commits behind head on master.

Files with missing lines	Patch %	Lines
src/layer/multiheadattention.cpp	60.48%	49 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5908      +/-   ##
==========================================
- Coverage   95.37%   95.36%   -0.01%     
==========================================
  Files         818      818              
  Lines      268009   268431     +422     
==========================================
+ Hits       255610   255990     +380     
- Misses      12399    12441      +42     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.