Removed HEADER_V72 and HEADER_V73

[radarhere]

Suggestions for python-pillow#6487

Feel free to disagree with any of my rearranging of code.

• Rather than always defaulting the pixel format to RGBA8888 when saving, use that for RGBA, RGB888 for RGB, I8 for L and IA88 for LA.
• Rather than writing the thumbnail image to a BytesIO buffer to get the length before writing it, _get_texture_size() can be used to get the length.
• Comment unused VtfPF enum values. I presume that's the pattern you were going for with the commented values.
• Removed HEADER_V72 and HEADER_V73. It seems like duplicate code to me, so instead of packing the entire header at one, just pack the HEADER_V70 portion, and then append the rest.
• Convert to RGBA for low-res DXT1, rather than RGB. I don't imagine there's any reason to abandon the alpha channel
• Thumbnail does not scale up, so thumb.thumbnail((min(16, width), min(16, height))) can be simplified to thumb.thumbnail((16, 16))
• DXT1 supports 8-bit alpha, so the CompiledVtfFlags.EIGHTBITALPHA flag can be used there as well
• Only CompiledVtfFlags.EIGHTBITALPHA for bcn encoding if there is an alpha channel present

[radarhere]

Cherry-picked into the PR.