[nihalpasham]

What this adds

The book page on kernels and device functions gets an "Inlining helpers" subsection and a one-page figure for the inline intent mapping that #1511 landed: #[inline] as inlinehint, #[inline(always)] as alwaysinline, #[inline(never)] as noinline, and the collector-boundary exception for generic #[device] helpers. Every quoted line in the figure comes from the inline_intent example built on an RTX 5090.

before: the page explains #[kernel] and #[device] and says a helper is "typically inlined"; nothing about #[inline]
after:  a short subsection with the three mappings, the one exception, and the figure below

What we fixed

• The inline_intent/Cargo.lock committed with feat(cuda-oxide): carry inline intent through the device pipeline #1511 predates the Tile libraries: call Rust device functions from Tile kernels #1545 lock re-alignment, so the first build of the example added four entries and left the tree dirty. A resolve against the existing lock adds itoa 1.0.18, oxide-tile-export 0.2.1, serde_json 1.0.154 and zmij 1.0.23, the versions the other 239 locks carry.

Verification

Checked at 128e5c478 (main 717432490 plus these two commits):

• ✅ just book builds with warnings as errors (the {figure} directive resolves; Sphinx reports no warning).
• ✅ The SVG renders in light and dark through headless Chrome; no line overflows its panel; no em or en dash in the file.
• ✅ scripts/sync-example-locks.sh --check: 240 example locks agree on one version per crate, and the inline_intent lock now lists the same packages as vecadd's apart from the two example crates themselves.

Caveats

• Docs and a lockfile only; no compiler code changes.

[nihalpasham]

/ok to test 128e5c4