vello_common: Compute rect edge-row tile alphas once per row

[nicoburns]

Summary

vello_common::rect::render computed per-pixel float coverage separately for every tile in a rect's top and bottom strip rows. Profiling cached glyph rendering in vello_cpu (each cached glyph is drawn as a rect fill, so ~9k small rects per frame) showed rect::render at ~20% of single-threaded time.

The top and bottom rows only have three distinct tile shapes: left edge, right edge, and interior. This change computes the alpha for each of those once per row and vectorises the calculation. Output is bit-identical.

Part 1 of 3 of the CPU glyph-cache speedups. It is independent of the other two:

• the image-resolve cache in Fine (#)
• the alpha-mask fast path

Measurements

Throwaway probe: 1600×1200 target, ~9.3k glyphs, 14px, atlas_cache(true), mean of 60 warm frames on a noisy Linux VM. Scene encode time for the cached run, main → this branch:

threads	main	this branch
1	5.69 ms	3.46 ms
4	4.19 ms	3.41 ms
8	4.64 ms	3.35 ms

cargo test -p vello_cpu -p vello_common passes.

Link to Devin session: https://dioxus.staging.devinenterprise.com/sessions/55344a6e161d48b4a3a8688fabe11d4c
Open in Devin Desktop: https://dioxus.staging.devinenterprise.com/desktop/session/55344a6e161d48b4a3a8688fabe11d4c?variant=devin-insiders
Requested by: @nicoburns

[staging-devin-ai-integration]

I'll fix CI failures and address comments from users with write access that start with 'Devin'.

• Disable automatic comment, CI, and merge conflict monitoring