Modify log_errors context manager to avoid traceback leaks

[sjperkins]

In the existing implementation, tracebacks from exceptions aren't properly garbage collected.
The strategy in this PR avoids this.
Tested in #6281 when enabling previously disabled check_instance tests.

• Tests added / passed
• Passes pre-commit run --all-files

[crusaderky]

Suggested change

	i = self.unroll_stack
	while traceback.tb_next and i > 0:
	traceback = traceback.tb_next
	i -= 1
	for _ in range(self.unroll_stack):
	if not traceback.tb_next:
	break
	traceback = traceback.tb_next

[github-actions]

Unit Test Results

       16 files  ±0         16 suites  ±0   7h 34m 33s ⏱️ - 11m 3s
  2 767 tests ±0    2 688 ✔️ +2       78 💤  - 2  1 ❌ ±0 
22 098 runs  ±0  21 070 ✔️  - 6  1 020 💤  - 1  8 ❌ +7 

For more details on these failures, see this check.

Results for commit 50814e0. ± Comparison against base commit 8411c2d.

[crusaderky]

ERROR    test_utils:utils.py:788 err4
Traceback (most recent call last):
  File "/home/runner/work/distributed/distributed/distributed/utils.py", line 765, in wrapper
    return func(*args, **kwargs)
  File "/home/runner/work/distributed/distributed/distributed/tests/test_utils.py", line 862, in inner
    raise CustomError("err4")
test_utils.test_log_errors.<locals>.CustomError: err4

I'm a bit worried that outer is nowhere to be found in the stack trace - and also that the output is now on stderr.
The use case that err4 replicates is a very common one in our codebase.
Could you find out what's happening?
Please let me know if you get stuck.