Writing TIFF tags: improved BYTE, added UNDEFINED

[radarhere]

Resolves #3677

This PR has two changes.

Firstly, Pillow currently treats BYTE tags of count N as if they might have N strings. I do not believe this should be the case. To look at count values for character-related tags, the BYTE tag GPSAltitudeRef has count 1 for 1 character, the ASCII tag GPSDateStamp has count 11 for 11 characters, and the UNDEFINED tag Flashpix Version has count 4 for 4 characters. So, I think BYTE tags of count N have N characters, and are only single strings.

Secondly, this PR changes the following current code, to no longer force BYTE tags into ASCII tags.

Pillow/src/encode.c

Lines 732 to 736 in d23df72

	if (PyBytes_Check(value) &&
	(type == TIFF_BYTE || type == TIFF_UNDEFINED)) {
	// For backwards compatibility
	type = TIFF_ASCII;
	}

This second change was made because in trying to resolve the issue, I found that libtiff was throwing an error for the XMLPACKET tag. It seems reasonable that it would - it is a BYTE tag, and we are instead passing ASCII.

Together, these changes resolve the issue.

[radarhere]

I've added another commit to this PR, since they both involve the same code, to resolve #4252. Pillow has been just been skipping UNDEFINED tags when writing with libtiff. This commit adds writing of UNDEFINED tags.