[analyzer] Break dependency between analyzer message codegen and analyzer.

This fixes a bootstrapping problem: previously we could get into a
situation where the generated files for analyzer diagnostic messages
were out of date, but the code generator for those messages couldn't
be run, because it imported the analyzer itself, which contained
compiled errors due to the generated files being out of date.

There were two problematic dependencies:

- The `GeneratedContent.check` and `GeneratedContent.checkAll` methods
  relied on invoking the Dart formatter as a library (rather than
  invoking `dart format` as a subprocess); the Dart formatter library
  imports the analyzer. Fortunately, this functionality is not needed
  by the code generators themselves. It's only needed by the _tests_
  that verify that code generation has run. So I was able to break the
  dependency by moving this functionality into a new file,
  `package:analyzer_utilities/generated_content_check.dart`, which is
  imported only by the code generation tests. The new API is slightly
  different: `GeneratedContent.checkAll` has been replaced by an
  extension method `check` on `Iterable<GeneratedContent>`.

- The diagnostic message code generator used to have a post-processing
  step that made use of the scanner to find error messages used by the
  parser that needed to be translated, and issued a warning if it
  found any. This check was useful a long time ago, when we were
  unifying the analyzer and CFE parsers, and is no longer useful, so
  I've removed it.

Change-Id: I509b741faca3b9ae21c23936001105ded731b490
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/442241
Reviewed-by: Konstantin Shcheglov <scheglov@google.com>
Reviewed-by: Samuel Rawlins <srawlins@google.com>
Commit-Queue: Paul Berry <paulberry@google.com>