CoverageException is fine as a base class, but not good to use for
raising (and catching sometimes).  Introduce specialized exceptions that
allow third-party tools to integrate better.

The old code expected a dict with int keys, but iterables are simpler, so get
rid of a lot of dict.fromkeys() nonsense.

.. and also add trailing commas where needed to remove ambiguity.

I used semgrep with this config:
```
rules:
  - id: implicit-string-concatenation
    pattern: |
      "$X" "$Y"
    languages:
      - python
    message: string cat BAD
    severity: INFO
```

Running it showed this:

```
$ semgrep --config=../lab/string-cat.yml coverage tests doc ci igor.py setup.py __main__.py
Running 1 rules...
setup.py
severity:info rule:lab.implicit-string-concatenation: string cat BAD
128:            'https://tidelift.com/subscription/pkg/pypi-coverage'
129:            '?utm_source=pypi-coverage&utm_medium=referral&utm_campaign=pypi'

tests/test_api.py
severity:info rule:lab.implicit-string-concatenation: string cat BAD
319:            "!coverage.py: This is a private format, don't read it directly!"
320:            '{"lines":{"/private/tmp/foo.py":[1,5,2,3]}}'

tests/test_html.py
severity:info rule:lab.implicit-string-concatenation: string cat BAD
614:            ('<span class="key">if</span> <span class="num">1</span> '
615:             '<span class="op">&lt;</span> <span class="num">2</span>'),
--------------------------------------------------------------------------------
616:            ('    <span class="nam">a</span> '
617:             '<span class="op">=</span> <span class="num">3</span>'),
--------------------------------------------------------------------------------
665:            ('<span class="key">if</span> <span class="nam">x</span> '
666:             '<span class="op">&lt;</span> <span class="num">2</span>'),
--------------------------------------------------------------------------------
667:            ('    <span class="nam">a</span> <span class="op">=</span> '
668:             '<span class="num">3</span>'),
--------------------------------------------------------------------------------
949:            ('<span class="key">if</span> <span class="num">1</span> '
950:             '<span class="op">&lt;</span> <span class="num">2</span>'),
--------------------------------------------------------------------------------
951:            ('    <span class="nam">a</span> <span class="op">=</span> '
952:             '<span class="num">3</span>'),
ran 1 rules on 134 files: 8 findings
```

This wasn't all of the instances, so I wrote a bug report: https://github.com/returntocorp/semgrep/issues/4184

This makes coverage warnings visible when running test suites under
pytest.  But it also means some uninteresting warnings would show up in
our own test suite, so we had to catch or suppress those.