Add BlurredRoundedRectangle

[PoignardAzur]

Copy-pasted from vello_common.

[waywardmonkeys]

This is an interesting one ...

In anyrender, it transports the transform along with it:

pub struct BoxShadowCommand {
    pub transform: Affine,
    pub rect: Rect,
    pub brush: Color,
    pub radius: f64,
    pub std_dev: f64,
}

In imaging, it does as well and includes a composite:

/// A solid-color rounded rectangle blurred with a gaussian filter.
#[derive(Copy, Clone, Debug, PartialEq)]
pub struct BlurredRoundedRect {
    /// Geometry transform.
    pub transform: Affine,
    /// Unblurred rectangle bounds.
    pub rect: Rect,
    /// Solid color used by the blurred rectangle.
    pub color: peniko::Color,
    /// Uniform corner radius in user-space units.
    pub radius: f64,
    /// Gaussian standard deviation in user-space units.
    pub std_dev: f64,
    /// Per-draw compositing.
    pub composite: Composite,
}

Obviously, we could use this to store the 4 fields as a single shared structure definition, but that wouldn't necessarily make the API better / different.

Not sure.

[xorgy]

@waywardmonkeys I could see that being useful/convenient for serializing or for lowering into renderers without the specialization (likely a straight copy, rather than a manual swizzle). In that sense it would be ‘better’ in at least that way.

[PoignardAzur]

I think vello_common's version makes the most sense for peniko.

• First, it's easier to compose than to un-compose, and peniko's items should lean towards composition. It's better to have importers pack a lean BlurredRoundedRectangle in a tuple with other fields, that to have them pack an extended type and ignore some of the fields.
• Second, this is coherent with other peniko types. peniko::Gradient doesn't include a transform or a blend mode.

[PoignardAzur]

I would like to request we merge this without further delay.

Assuming the lack of a transform is the only objection, I honestly don't think it's an objection strong enough to stall this PR for three months. I think the real reason is that nobody has had the bandwidth to review this, but also... It's a twenty-line copy-paste from an existing Linebender project? I'm fine with a rubber-stamp review.