//! Widgets: named assemblies of things the description already says.
//!
//! The third tier, between a primitive and `Region::Bespoke`. A primitive is
//! drawn from scratch by every renderer and has to be something every host can
//! honour; a bespoke region is one app's own and carries no contents the
//! description knows. A widget is neither: a *name* over an assembly of members
//! the vocabulary already has.
//!
//! Full argument: wiki `widget-tier`.
//!
//! # What is left here
//!
//! Glue, and only glue. The widgets themselves live in `quasi-basics`, which is
//! where the carousel went once it had been proved against three real pages.
//! Each module here is an Askama entry point: it turns whatever the templates
//! happened to need into the widget's own types, renders the fragment, and
//! stops.
//!
//! A widget that has never rendered anything is a guess about what the assembly
//! needs, so a new one is welcome to start here and leave later. The carousel
//! found a missing primitive on its way through
//! (`makeover_layout::Image`, 0.21.0), which is the argument for proving one in
//! an app before it becomes shared.
//!
//! # A primitive's glue is a sibling of this directory, never a member
//!
//! Ruled by Max 2026-08-21 (`27d5e5b8`), and written here because this doc is
//! what gets read when someone asks where a described fragment goes.
//!
//! **A described fragment inside an Askama page is legitimate at any
//! granularity the vocabulary already names, primitive included.** One `Field`
//! qualifies; so does one `Act`. The question was whether that made a single
//! primitive a widget, and the answer is that it does not — it makes it a
//! primitive with an entry point. So the glue module sits at `src/quasi/`,
//! beside `rich_field.rs`, `upload_field.rs` and `export_act.rs`, and this
//! directory keeps meaning what the paragraphs above say it means: a name over
//! an assembly. The "misuses the word widget" objection is answered by
//! placement rather than by declining the pattern.
//!
//! What it buys is that a shared control converts at its own granularity
//! instead of once per parent. `export_act.rs` is the first one taken: five
//! Export CSV buttons in four templates, described once. The same applies to
//! the shared act rows — `partials/link_row.html`, `partials/tag.html`, and the
//! two identical delete buttons in `tabs/project_content.html` and
//! `tabs/project_blog.html`.
//!
//! ## What it costs, stated so it is not rediscovered
//!
//! A live visual change on a page no flag gates. The templates write
//! `class="btn-secondary text-sm"` and a described act emits `class="button"`,
//! so the control lands on the design system's button rather than the app's:
//! `--surface-raised` and `--bevel-raised` where `.btn-secondary` sets
//! `--surface-sunken`, `--shadow-raised` and its own padding and font size.
//! That is the swap, it is deliberate, and it wants an eyeball rather than a
//! test.
//!
//! It is not, any more, a loss of the design system's button altogether.
//! `35756077` measured that `style.css` is unlayered and outranks
//! `@layer makeover` at any specificity, so a bare `button { }` rule was taking
//! the surface, the bevel **and the tone** off every described act. Fixed by
//! the `revert-layer` handoff at `style.css:473`. A described act that reaches
//! for `Tone::Danger` gets it now, which it did not before 2026-08-21.

pub mod carousel;