support batch inference for crnn and segocr

[cuhk-hbsun]

support batch inference for all the resize modes in text recognition algorithms, including:

1. Resize to a fix shape;
2. Keep aspect ratio resize and pad to a fix shape;
3. keep aspect ratio resize, shape is not fixed.

So that crnn and segocr (keep aspect ratio resize) can use batch inference during testing.
(Note: There may be a mismatch in eval metric for batch_size=1 and batch_size>1)

[codecov]

Codecov Report

Merging #407 (4219569) into main (a2af2b4) will decrease coverage by 0.08%.
The diff coverage is 83.33%.

❗ Current head 4219569 differs from pull request most recent head 1b6c11d. Consider uploading reports for the commit 1b6c11d to get more accurate results

@@            Coverage Diff             @@
##             main     #407      +/-   ##
==========================================
- Coverage   83.21%   83.12%   -0.09%     
==========================================
  Files         137      137              
  Lines        9155     9162       +7     
  Branches     1299     1301       +2     
==========================================
- Hits         7618     7616       -2     
- Misses       1244     1252       +8     
- Partials      293      294       +1     

Flag	Coverage Δ
unittests	83.12% <83.33%> (-0.09%)	⬇️

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

Impacted Files	Coverage Δ
mmocr/apis/inference.py	75.64% <ø> (-1.47%)	⬇️
mmocr/core/visualize.py	91.10% <0.00%> (-0.66%)	⬇️
mmocr/models/textrecog/convertors/seg.py	91.78% <100.00%> (+2.89%)	⬆️
...s/textrecog/recognizer/encode_decode_recognizer.py	91.95% <100.00%> (+0.59%)	⬆️
...mocr/models/textrecog/recognizer/seg_recognizer.py	89.55% <100.00%> (+0.48%)	⬆️
mmocr/datasets/pipelines/custom_format_bundle.py	88.23% <0.00%> (-5.89%)	⬇️
mmocr/datasets/pipelines/transforms.py	80.36% <0.00%> (-1.16%)	⬇️

---

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 a2af2b4...1b6c11d. Read the comment docs.

[gaotongxiao]

Batch inference hurts the performance of both models. Users should be clearly warned in the docs.