feat(pointer,rich_text): a hand that closes on click, and pill spans

[LeadcodeDev]

Closes #381. Closes #382. Refs #388.

#381 — a hand glyph that closes on click

PointerGlyph (arrow default, hand, grab) plus glyph and click_glyph. glyph_path() is one match arm dispatching to three path builders, mirroring how colors() already dispatches fill and outline — the same shape, not a parallel mechanism.

It reuses the existing click signal. active_glyph() consumes click_progress, the same Option that already drives the click ring, so there is no second clock to keep in sync. A path waypoint clicking on arrival closes the hand for exactly as long as it rings.

Both new glyphs put their reference point at local (0, 0) like the arrow's tip, so path and click_at coordinates mean the same thing whichever glyph is chosen. grab is the fist alone, so closing removes exactly the finger's pixels.

Red proof:

before any click the open hand must paint its extended finger (white fill),
got (17, 24, 39, 255)
  left:  (17, 24, 39, 255)
  right: (255, 255, 255, 255)

#382 — pill spans in rich_text

RichTextSpan gains background, padding, border-radius and rotation.

The padding affects layout, which is the part that is easy to get wrong. A post-wrap reflow pass adds padding.left to a pill run's first token and padding.right to its last, cumulatively, so following spans shift over. RichTextIntrinsic calls compute_layout, so the node's measured size agrees with what is painted without a second code path.

Red proof:

a pill's horizontal padding must be added to the line's advance, moving the
following span over by left+right padding:
plain x=66.71875, pill x=66.71875, expected shift 32

Wrapped pills get box-decoration-break: clone — grouping is per rendered line, not per source span, so a pill that wraps gets its own padded box per fragment instead of one box straddling two lines. The issue asked for it literally and there is a dedicated test.

Rotation turns the box and its glyphs together about the pill's own centre, inside a saved canvas state — layout is never touched, which its test asserts alongside the pixel check.

Verification

Eight new tests (four per issue), each proven red by reverting the fix. cargo fmt --all --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test --workspace (1640) all clean. Comment count still 0.

Three things to know

• click_glyph without glyph: "hand" is legal but inert — nothing ever shows a resting grab. Documented rather than rejected.
• A pill's wrap decision is made on unpadded text width, so a pill sitting right at the boundary can overflow slightly instead of wrapping a word earlier. Documented as a known approximation.
• box_builder.rs was touched outside the stated ownership: its two hand-written RichTextSpan literals are the only other exhaustive construction sites in the tree, and adding four fields broke them. Four Nones, no logic. Flagged rather than buried.

Written comment-free, per the codebase-wide rule from #345.