Improved handling of PPM header

[Piolie]

Addresses #5104. That discussion didn't get any replies, so I decided to open this PR to make my intentions more explicit.

Changes proposed in this pull request:

• remove some dead code;
• handle non-decimal ASCII in headers properly;
• handle comment strings that can start in the middle of a token and end with LF or CR;
• make the exception types raised more consistent (although the resolution of issues raised in Add proper exception types #2339 and related is still pending);
• add two tests; change the exception type expected in other two.

[gofr]

I'm just a random contributor so I don't dare to "Approve", but apart from the minor notes I added this looks good to me. 👍

[Piolie]

I hurried to push and forgot to lint and a raise from None. I'll fix that and the other tests as noted by @gofr.

[radarhere]

I'm not quite on board with this one. The idea here is that b"\x00" is not caught by

Pillow/src/PIL/PpmImagePlugin.py

Lines 57 to 58 in 92c26a7

	if c > b"\x79":
	raise ValueError("Expected ASCII value, found binary")

even though \x00 is not greater than \x79, and only for it to be raised as a different ValueError later when it turns out that b"128\x00" is not an integer.

There isn't a valid scenario where the token/magic needs to be something other than a number or a letter. I've pushed a commit to change this test into a check that a ValueError is raised when a token is not an integer.