Drop support for EOL Python 3.5

[hugovk]

For #4764.

Continuation of #4746.

Changes proposed in this pull request:

• Update Black target to py36
• Upgrade Python syntax for 3.6+ using pyupgrade `git ls-tree -r HEAD --name-only | grep ".py$"` --py36-plus and manually

For clarity, I left some as percent formatters:

    print("Size: %sx%s" % im.size)
...
            raise OSError("Unsupported BMP Size: (%dx%d)" % self.size)
...
            imf.save("out-%s-%s-%s.png" % size)
...
            page_contents = b"q %d 0 0 %d 0 0 cm /image Do Q\n" % (
                int(width * 72.0 / resolution),
                int(height * 72.0 / resolution),
            )

And as .format:

        print(
            "{}: {:.4f} s  {:.6f} per iteration".format(
                label, endtime - starttime, (endtime - starttime) / (x + 1.0)
            )
...
                    tga_path = "{}_{}_{}.tga".format(
                        path_no_ext, origin, "rle" if rle else "raw"
                    )

Happy to change any of these too if we think they're better, and if I missed any.

Can also split the PR by, say, src and Tests directory if it makes it easier to review. I recommend marking files as viewed
.

[nulano]

There are a few places where you can use {var!r} instead of {repr(var)}, and a few more small issues.

[nulano]

Suggested change

	raise OSError(f"Unsupported header size {repr(header_size)}")
	raise OSError(f"Unsupported header size {header_size!r}")

[hugovk]

Thanks for detailed review!

https://stackoverflow.com/a/44800859/724176 says of !r:

It just calls the repr of the value supplied.

It's usage is generally not really needed with f-strings since with them you can just do repr(self.radius) which is arguably more clear in its intent.

!r (repr), !s (str) and !a (ascii) were kept around just to ease compatibility with the str.format alternative, you don't need to use them with f-strings.

I'm tempted to keep the more explicit version, and the f-string PEP says "!s, !r, and !a are redundant".

Maybe we should replace the other handful of existing !rs with repr.

[hugovk]

Maybe we should replace the other handful of existing !rs with repr.

Done in last push.

[nulano]

There's one more in Tests/test_imagemath.py at the top. I personally prefer !r for brevity, but it's fine as long as it's consistent. Perhaps it would be a good idea to add a lint rule for this?

[hugovk]

Found another little bit to remove.

[hugovk]

Will merge in a bit if there's no further comments!