Convert synchronization dependency to data dependency

[JoongunPark]

Summary

This update is for considering synchronization dependency we can extract from Holistic Trace Analysis (HTA, https://github.com/facebookresearch/HolisticTraceAnalysis)

It uses new input field for Chakra converter, 'sync_dep'.
Required updates are developing on forked repo (https://github.com/JoongunPark/param/tree/refactor)
The updates will be sent as PR to Param (https://github.com/facebookresearch/param) or Taekyung's forked repo (https://github.com/TaekyungHeo/param/tree/refactor)

Test Plan

Command haven't changed.

# Install package from source
$ pip install .

# Run converter 
python3 -m chakra.et_converter.et_converter\
    --input_type PyTorch\
    --input_filename ./result/rank_0.json\
    --output_filename ./result/rank_0.et\
    --num_dims 1

Tested input from Resnet-50 with GTX1070 (rank 0)
rank_0.json

Tested output (Zipped)
rank_0.zip

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[JoongunPark]

Close this PR since I made another PR for the later Chakra version.