feat: let styles other than `grid` render names and roles

[hahwul]

We wanted maintainer roles ("Creator, Leader" / "Leader") on a voronoi wall. Those two requirements turn out to be mutually exclusive.

Grid#draw_block gates the role <text> behind show_names?:

if grid.show_names?
  label(io, name_label(user), center, y + grid.avatar_size + 13)
  if role = role_label(user)
    io << %(    <text ... font-size="9" ...>#{SVG.escape(role)}</text>\n)
  end
end

Every other renderer routes the role into title_for, i.e. a <title> tooltip. The README is upfront that a README embed renders as <img> where tooltips are inert — but the consequence is that "label your maintainers" and "pick a style other than grid" cannot both be satisfied, and the canonical embed is exactly the one where tooltips do not work.

We ended up choosing the style and dropping the labels, keeping role: in the config purely so the metadata survives for a future change.

Suggestion: an optional caption layer any style can opt into — a reserved strip beneath the block listing the labelled users, or a small badge in the cell corner for users that have a role. Even a grid-only-today labels: below that other styles could adopt later would unblock this.

Related: #4 (per-user emphasis) — labelling and sizing are the two ways to say "this person is a maintainer", and neither is currently available outside grid.

[hahwul]

Leaving this open, but with a concrete shape, since it came up again while clearing the
rest of the batch.

The constraint is real and worth restating plainly: role is only ever rendered by
grid, and only when show_names is on. Every other renderer routes it into
title_for, i.e. a <title> tooltip — and the canonical README embed is an <img>,
where tooltips are inert. So "label your maintainers" and "pick a style other than
grid" genuinely cannot both hold today.

The cheap version: captions belong to the base class, not the styles

Nothing about a caption needs to know how a style places avatars. Renderer#render
already sizes every section, stacks them, reserves TITLE_HEIGHT for group titles, and
owns the theming — it can reserve a strip under a block the same way, and every style
gets labels without touching draw_block.

That is the property worth aiming for: adding this should not mean seven renderer
changes, and it should not mean a style added later has to opt in again.

Sketch:

labels: roles          # none (default) | roles

• none — today's behaviour exactly. grid's show_names keeps working as it does now.
• roles — the base class reserves a strip beneath each section and lists the users in
  it that have a role, as Name — Role, in the existing label/role palette colours.

Only labelled users appear, which is what keeps it useful: on a 24-person wall the two
maintainers are the whole caption, and the strip is one line. It also degrades sensibly —
a section where nobody has a role reserves no strip.

Text width is already estimated by text_width for exactly this reason (widening the
canvas so grid labels are not clipped), so wrapping to the block width is tractable with
what is there.

What I would not do

A badge in the cell corner. It needs per-style geometry — voronoi cells are irregular
and a corner may be a wedge the clip path throws away, which is the same problem cover
already solves for avatars. Much more work, worse failure mode, and it does not
generalise to a style added later.

Ordering

This is independent of #4 and smaller. The two together are the "mark a maintainer"
story — size and label — but a caption strip is useful on its own, and unlike per-user
scale it has an honest interpretation in every style including the fixed lattices
(grid, honeycomb, stencil), which is where scale has none.

So if only one of the two gets built, this is the one I would pick.