Make the static error test updater run the analyzer as a library.

Prior to this change, the static error test updater would spawn a new
analyzer CLI for each updated test file. This was incredibly slow, like
5-10 seconds for each file.

This change instead uses the analyzer as a library. It creates a single
analysis context collection for all of the files to be updated and then
gets the errors from that.

On my machine, it's about 100x faster. It's fast enough to routinely
update *all* of the test expectations in a batch. (Or, at least, it will
be if we can make a similar change for CFE.)

Change-Id: I39059489e966abf040e9f995505c7013b18d4930
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/396574
Reviewed-by: Konstantin Shcheglov <scheglov@google.com>
Reviewed-by: Paul Berry <paulberry@google.com>
Commit-Queue: Konstantin Shcheglov <scheglov@google.com>
Auto-Submit: Bob Nystrom <rnystrom@google.com>