[Fix] Fix random seed

[gaotongxiao]

Motivation

The previous implementation did not always force all the processes to have the same random seed in DDP training, which may lead to inconsistent results across different runs. This PR fixes this issue by always broadcasting a random seed to all the processes during the training initialization phase.

Reference

open-mmlab/mmdetection#6457

[codecov]

Codecov Report

Merging #600 (0e97f2c) into main (284b5ac) will increase coverage by 0.00%.
The diff coverage is 21.42%.

@@           Coverage Diff           @@
##             main     #600   +/-   ##
=======================================
  Coverage   84.86%   84.86%           
=======================================
  Files         143      143           
  Lines        9802     9816   +14     
  Branches     1417     1420    +3     
=======================================
+ Hits         8318     8330   +12     
- Misses       1157     1158    +1     
- Partials      327      328    +1     

Flag	Coverage Δ
unittests	84.86% <21.42%> (+<0.01%)	⬆️

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

Impacted Files	Coverage Δ
mmocr/apis/train.py	16.66% <21.42%> (+1.04%)	⬆️
mmocr/datasets/pipelines/transforms.py	83.99% <0.00%> (+1.48%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 284b5ac...0e97f2c. Read the comment docs.