[Fix] no-unused-prop-types: false positive with callback

[golopot]

Fixes #2350

This pr fixes getParentStatelessComponent when the current traversed node is like a callback (is a function and is an argument). The function getParentStatelessComponent is revised to use a more whitelist-ish approach.

Fixes false positive like:

function Foo(props) {
  useEffect(() => {
    const { a } = props;
    document.title = a;
  });
  
  return <p/>;
}
  Foo.propTypes = {
  a: PropTypes.string,
}

[golopot]

These cases are

const Foo = () => <p />;

module.exports = function Foo() {};

module.exports = {
  Foo() {},
};

function hof() {
  return function Foo() {};
}

export default function Foo() {}

[ljharb]

Can we add explicit test cases for each of them?

[golopot]

Currently the tests cases for them are scattered in several rules, do you want a set of test cases centralized in, say, tests/rules/prop-types.js?

[ljharb]

it'd probably be ideal for every rule that uses component detection to have all of these cases, tbh

[golopot]

The tests are broken by the merge of #2184, which added a test case: https://github.com/yannickcr/eslint-plugin-react/blob/7bc55ccb6ed2c32d381f09d2f43a77ca60afaae3/tests/lib/rules/no-multi-comp.js#L345-L349

Do we want to support components defined in comma operator? (IMO, no.)

[ljharb]

Yes, we should support components in any expression position.