Honor Image `variant` size hint (icon/avatar vs feature/header placeholders)

[joestump-agent]

Part of #54.

Problem

renderImage (render/media.go:12-21) reads only Description and URL; the basic catalog's variant size hint (icon, avatar, smallFeature, mediumFeature, largeFeature, header) and fit are silently dropped. Everything renders as the same one-line 🖼 placeholder, so an agent asking for a small inline icon and one asking for a header image get identical output.

Fix (minimum honest version)

• icon / avatar → compact inline glyph + description, no full-width placeholder line — suitable for embedding in a Row.
• smallFeature / mediumFeature / largeFeature / header → keep the placeholder line; optionally scale placeholder height (1/2/3 lines) to convey the hint.
• fit remains ignored (meaningless without real image rendering); document that in the wire-format hint audit rather than dropping it silently.
• Media placeholder styling should also move onto the Placeholder slot from Expand render.Styles into a complete token set (component slots + Glyphs) #56 (today it reuses Caption).

Terminal graphics protocols (kitty/sixel/iTerm2) are explicitly out of scope for this issue — this is about making the placeholder honor the semantic hint.