Allow providing a text alternative/caption for all media/non-text-content

Suggestion

I would like to specify an alternative text for all media content uploaded to Matrix including casual uploads, user avatar and room avatar, and I think this should be supported to comply with WCAG 2.1 non-text content. I think that especially in the context of file uploads during chat (as in not avatar change) there should be an option to always have it visible.

Examples:

• The Matrix experience naturally improves for an vision impaired user.
• Terminal user interface applications could possibly show the alt-text where they cannot show an avatar?
• Matrix Suomi is using the image below as an avatar. Just like I can describe it here on the web, I would like to be able to describe it for users who cannot see the image itself, e.g. "Matrix Finland logo combines the brackets from Matrix logo with the cross as seen in the flag of Finland."
  • 
• LGBT* flags are very diverse and I imagine not even all LGBT* people recognize all the flags under the umbrella. When one is used as a room avatar or similar, it could be named and possibly a bit of where it comes from explained. This could also be used for other less known symbols.
• A furry could explain their fursona in greater detail than may be obvious from the text, especially in case of a less common "species".
• WCAG 2.1 has additional details for Understanding Success Criterion 1.1.1: Non-text Content
  .

This is similar to matrix-org/matrix-spec#882 (media attribution fields) and I was discussing this with @Sorunome, but she pointed out that attribution is different from text explanation what is in the picture and the user may not understand to put attribution/copyright information to this field.

Start by reading the linked WCAG 2.1 non-text content guidance and the related matrix-org/matrix-spec#882 issue. Determine how alt text would apply to casual uploads, user avatars, and room avatars, including when it should be visible; the work is done when the specification defines an agreed protocol model and behavior for these cases.