Added reading of earlier ImageMagick PNG EXIF data

[radarhere]

Resolves #4460

Before 'eXIf' chunks became part of the PNG specification, ImageMagick stored EXIF data in a zTXt chunk. As a zTXt chunk, it is read into Pillow's info dictionary as "Raw profile type exif".

So this PR suggests reading that data as a fallback, if an eXIf chunk is not present.

Installing ImageMagick-7.0.6-0, I ran convert over Tests/images/exif.png and created an image with the zTXt chunk.

When reading, I skip the first two lines - see the corresponding ImageMagick code https://github.com/ImageMagick/ImageMagick/blob/9caa8b0c4269163def2ce0fa6ff1754ec0c234d2/coders/png.c#L7911-L7921

[hugovk]

We could use pytest's parametrize on this test.

Before

    def test_exif(self):
        for test_file in (
            "Tests/images/exif.png",  # With an EXIF chunk
            "Tests/images/exif_imagemagick.png",  # With an ImageMagick zTXt chunk
        ):
            with Image.open(test_file) as im:
                exif = im._getexif()
            assert exif[274] == 1

After

    @pytest.mark.parametrize(
        "test_file",
        [
            "Tests/images/exif.png",  # With an EXIF chunk
            "Tests/images/exif_imagemagick.png",  # With an ImageMagick zTXt chunk
        ],
    )
    def test_exif(self, test_file):
        with Image.open(test_file) as im:
            exif = im._getexif()
        assert exif[274] == 1

One benefit is it (eg. pytest Tests/test_file_png.py::TestFilePng::test_exif) is run as two tests instead of one, and separates each test case so that one doesn't inadvertently depend on the other. And if the first fails, the second will still be run.

[radarhere]

Sounds reasonable. Done.

[hugovk]

Thanks!