This makes the unit tests more concise and easier to
follow.  Removes a lot of the duplication.  Also makes it
easier to write new filter tests.

The pattern is evaluated against the entire bucket.  This doesn't
mattern for suffix searches, but for a prefix search, you'd have to
include the bucket name.  This is inconsistent with filtering local
files.  Now they're both consistent.  Given:

  --exclude 'foo*'

This will filter any full path that startsw with foo.  So locally:

  rootdir/
    foo.txt      # yes
    foo1.txt     # yes
    foo/bar.txt  # yes
    bar.txt      # no

And on s3, we now have the same results:

  bucket/
    foo.txt      # yes
    foo1.txt     # yes
    foo/bar.txt  # yes
    bar.txt      # no

I also added debug logs to help customers troubleshoot why their
filters aren't working the way they expect.