Allow setting some Learner-related params via environment variables (or other Everett config options)

[AdeelH]

Overview

This PR

• adds support for the following env variables:
  • RASTERVISION_PREDICT_BATCH_SIZE - the batch size to use during prediction.
  • RASTERVISION_PREDICT_NUM_WORKERS - num_workers to pass to the PyTorch DataLoader during prediction.
• refactors code so that these env variables (as well as the previously added RASTERVISION_USE_ONNX) are accessed through RVConfig.
  • This means that these variables can also be set in the .rastervision profile.

Checklist

• Added needs-backport label if PR is bug fix that applies to previous minor release
• Ran scripts/format_code and committed any changes
• Documentation updated if needed
• PR has a name that won't get you publicly shamed for vagueness

Notes

N/A

Testing Instructions

N/A

[codecov]

Codecov Report

Merging #1978 (8f9fa23) into master (e56f93d) will decrease coverage by 0.01%.
The diff coverage is 86.36%.

@@            Coverage Diff             @@
##           master    #1978      +/-   ##
==========================================
- Coverage   86.20%   86.20%   -0.01%     
==========================================
  Files         193      193              
  Lines        9434     9451      +17     
==========================================
+ Hits         8133     8147      +14     
- Misses       1301     1304       +3     

Files	Coverage Δ
...ch_learner/rastervision/pytorch_learner/learner.py	82.66% <100.00%> (+0.12%)	⬆️
...vision_pipeline/rastervision/pipeline/rv_config.py	83.63% <81.25%> (-0.90%)	⬇️