[enhance]: Use RoIAlign implemented in MMCV for inference in CPU mode

[RunningLeon]

Since RoIAlign in mmcv.ops is supported with aligned=False in CPU mode, there's no need to use the op from torchvision in CPU mode.

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

Merging #3930 into master will decrease coverage by 0.04%.
The diff coverage is 88.00%.

@@            Coverage Diff             @@
##           master    #3930      +/-   ##
==========================================
- Coverage   62.27%   62.23%   -0.05%     
==========================================
  Files         221      221              
  Lines       16310    16305       -5     
  Branches     2755     2756       +1     
==========================================
- Hits        10157    10147      -10     
- Misses       5659     5666       +7     
+ Partials      494      492       -2     

Flag	Coverage Δ
#unittests	62.23% <88.00%> (-0.05%)	⬇️

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

Impacted Files	Coverage Δ
mmdet/models/dense_heads/paa_head.py	91.59% <ø> (+1.17%)	⬆️
mmdet/models/roi_heads/cascade_roi_head.py	71.19% <33.33%> (-0.60%)	⬇️
mmdet/apis/inference.py	21.51% <50.00%> (+0.78%)	⬆️
...mdet/core/bbox/iou_calculators/iou2d_calculator.py	96.77% <100.00%> (+8.31%)	⬆️
mmdet/models/losses/iou_loss.py	37.71% <100.00%> (-5.07%)	⬇️
mmdet/models/roi_heads/test_mixins.py	50.42% <0.00%> (-9.41%)	⬇️

---

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 502cee2...f687768. Read the comment docs.