add better error handling when tearing down service

[kptkin]

Fixes WB-10725

Description

What does the PR do?

When using wandb.teardown the system could be in a bad state. For example in a notebook setup where the service was closed, due to some kill signal, but token env variable is still present and the setup singleton was not removed. This PR adds better error handling for cases when some of the teardown operations could cause an exception and guarantee that the env variable is cleared to force reconnect on a new call to init.

Testing

How was this PR tested?

Add a unit tests that tries to create a bad state and see that the error was handled.

Checklist

• Include reference to internal ticket "Fixes WB-NNNN" (and github issue "Fixes #NNNN" if applicable)

[codecov]

Codecov Report

Merging #4161 (5912d81) into master (43e8ead) will increase coverage by 0.00%.
The diff coverage is 90.00%.

@@           Coverage Diff           @@
##           master    #4161   +/-   ##
=======================================
  Coverage   82.65%   82.66%           
=======================================
  Files         256      256           
  Lines       32541    32546    +5     
=======================================
+ Hits        26896    26903    +7     
+ Misses       5645     5643    -2     

Flag	Coverage Δ
functest	55.96% <70.00%> (+<0.01%)	⬆️
unittest	73.38% <90.00%> (+0.04%)	⬆️

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

Impacted Files	Coverage Δ
wandb/sdk/wandb_manager.py	94.07% <90.00%> (+0.22%)	⬆️
wandb/sdk/lib/sock_client.py	93.12% <0.00%> (-1.06%)	⬇️
wandb/sdk/wandb_run.py	91.16% <0.00%> (-0.19%)	⬇️
wandb/sdk/lib/git.py	79.01% <0.00%> (ø)
wandb/integration/tensorboard/monkeypatch.py	92.59% <0.00%> (+2.46%)	⬆️
wandb/sdk/internal/meta.py	90.79% <0.00%> (+3.06%)	⬆️