fix: dynamic convert onnx model to fp16 during start session

[OrangeSodahub]

Fix:

• conversion of onnx to fp16 will cover the original model file

Changes:

• move the converter from __init__() to start_session, using tmp_dir to do the conversion.
• Add test for onnx zip+fp16.

[codecov]

Codecov Report

Merging #876 (78fa087) into main (1fe3a5a) will decrease coverage by 11.37%.
The diff coverage is 100.00%.

@@             Coverage Diff             @@
##             main     #876       +/-   ##
===========================================
- Coverage   83.12%   71.74%   -11.38%     
===========================================
  Files          22       22               
  Lines        1529     1529               
===========================================
- Hits         1271     1097      -174     
- Misses        258      432      +174     

Flag	Coverage Δ
cas	71.74% <100.00%> (-11.38%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
server/clip_server/executors/clip_onnx.py	87.91% <100.00%> (+1.09%)	⬆️
server/clip_server/model/clip_onnx.py	87.30% <100.00%> (+20.63%)	⬆️
server/clip_server/executors/clip_tensorrt.py	0.00% <0.00%> (-94.60%)	⬇️
server/clip_server/model/clip_trt.py	0.00% <0.00%> (-85.72%)	⬇️
server/clip_server/model/trt_utils.py	0.00% <0.00%> (-83.52%)	⬇️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[ZiniuYu]

LGTM