x86 optimization for convolution int8 packed unified elempack

[nihui]

• opt sse2
• opt sse4.1
• opt xop
• opt avx2
• opt avx512
• opt vnni
• clear todo
• xop test
• drop old files
• restore non-packed path
• coverage++
• omp 16 args
• transpose write size 4
• aligned load

[codecov-commenter]

Codecov Report

Merging #4861 (29ed4af) into master (e8645e9) will increase coverage by 0.42%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #4861      +/-   ##
==========================================
+ Coverage   94.53%   94.96%   +0.42%     
==========================================
  Files         779      776       -3     
  Lines      221835   226111    +4276     
==========================================
+ Hits       209706   214718    +5012     
+ Misses      12129    11393     -736     

Impacted Files	Coverage Δ
src/layer/x86/convolution_packed_int8.h	97.22% <ø> (ø)
src/layer/arm/convolution_3x3_winograd.h	96.31% <100.00%> (ø)
src/layer/arm/convolution_3x3_winograd_bf16s.h	98.78% <100.00%> (ø)
src/layer/x86/convolution_x86.cpp	89.08% <100.00%> (-0.16%)	⬇️
src/layer/x86/x86_usability.h	100.00% <100.00%> (+4.24%)	⬆️

... and 49 files with indirect coverage changes