use task_name instead of task.name

[zphang]

There are two sources of task names in jiant.

1. The task-names provided in the task_name->task mapping from the task_config_dict
2. The task.name, which is derived from the task config.

In all parts of the jiant training/eval loop, we use task_name, including for creating the models. The one exception is the runner calling the model, where we use the task object (and hence task.name).

To resolve this conflict, we will assign the task.name = task_name and throw a warning if they do not agree.

[codecov]

Codecov Report

Merging #1224 into master will decrease coverage by 0.12%.
The diff coverage is 75.00%.

@@            Coverage Diff             @@
##           master    #1224      +/-   ##
==========================================
- Coverage   56.62%   56.49%   -0.13%     
==========================================
  Files         145      144       -1     
  Lines       10413    10371      -42     
==========================================
- Hits         5896     5859      -37     
+ Misses       4517     4512       -5     

Impacted Files	Coverage Δ
jiant/proj/main/components/container_setup.py	91.17% <75.00%> (-1.46%)	⬇️
jiant/shared/runner.py	75.00% <0.00%> (-6.09%)	⬇️
jiant/proj/main/runscript.py	72.41% <0.00%> (-0.63%)	⬇️
jiant/proj/simple/runscript.py	80.00% <0.00%> (-0.33%)	⬇️
jiant/proj/main/metarunner.py	81.74% <0.00%> (-0.08%)	⬇️
jiant/tasks/retrieval.py	100.00% <0.00%> (ø)
jiant/tasks/evaluate/core.py	36.17% <0.00%> (ø)
jiant/tasks/lib/quail.py

---

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 cb601cf...f1b1d04. Read the comment docs.