Update to pass under the latest flake8-bugbear

[sirosen]

Primarily, flake8-bugbear flags several of our exceptions as being
unsafe to stringify, copy, or pickle because they don't pass args to
super() in the mandated way. (This is the new B042 rule.)

In order to pass, most of these cases are simple to update to define and
test explicit str() and copy.copy() calls. Copy is a simpler
operation than pickling and exercises the same reducer. For the target
error types, defining custom __reduce__ methods is sufficient.

For GlobusAPIError, because it's so key to SDK interfaces, no action
is taken for now, other than the addition of a TODO comment.

New unit tests exercise str and copy.copy on all of the updated errors.

[sirosen]

I believe the codspeed "improvement" is not real, and is an artifact of some other change (e.g., new machine image from GitHub). I can't see any way that making this change would improve performance of that part of the code.

My current working theory is that changing the imports in the exception modules is enough to rejigger the layout of some data in memory and this benchmark is so hypersensitive that it detects the difference. That might point at a potential improvement to benchmarks for the future, but I'm not sure what exactly we should be doing to make this more stable.