[jarmak-personal]

Closes #52

Summary

• draw a focused set of 26 uniform light/heavy and rounded box-drawing glyphs as device-pixel-aligned Canvas geometry
• align cell metrics to the physical-pixel grid without shrinking the existing CSS-ceiled dimensions
• keep canvas backing dimensions stable at fractional DPRs and preserve selection, inverse, faint, and cursor styling
• retain font rendering for mixed-weight, dashed, double-line, block-element, and grapheme cases

Design notes

I reviewed upstream PRs coder#128 and coder#146, but did not copy either implementation: coder#128 expands into unrelated block/Powerline/visual-infrastructure work, while coder#146 later includes unrelated performance changes. This PR keeps the fix scoped to the reproducible box-drawing seams and uses a single fill path for intersecting straight strokes so translucent/faint cells composite once.

Verification

• bun run fmt
• bun run lint (pre-existing diagnostics only)
• bun run typecheck
• bun test (541 passing, 0 failing)
• bun run build
• geometry coverage across 11/15/19/21px fonts and DPR 1/1.25/1.5/2
• ESM and UMD external-WASM smoke checks

[jarmak-personal]

External review disposition:

• Gemini 3.6 Flash: CLEAN.
• GPT-5.4 Mini raised two faint-style behaviors (block-cursor accent alpha and decoration alpha). I traced both through origin/hvir-main; both code paths are unchanged on the base branch and are not regressions from this PR, so they are intentionally left out of Eliminate box-drawing gaps at fractional device scales #52.

No review-driven code changes were required.

[jarmak-personal]

Full Opus review disposition (claude-opus-4-8, max effort): approve / merge.

Opus found no blocking correctness, security, performance, or regression issue. It independently verified the device-grid invariant, stroke centering, single-composite faint behavior, run isolation, fallback/grapheme path, all 26 glyph mappings and rounded orientations, stable backing-store sizing, and test-harness fidelity.

Non-blocking notes consciously accepted for this scoped fix:

• 1/3-device-pixel light/heavy strokes prioritize crisp odd-width geometry over font-scaled/native weight parity.
• Rounding cells upward at fractional DPR can slightly reduce fitted rows/columns; this avoids shrinking legacy cell bounds or clipping fallback glyphs.
• Runtime DPR changes remain a pre-existing renderer lifecycle gap and belong in a separate follow-up.
• Broader mixed/dashed/double/block glyph coverage remains intentionally font-rendered.

No Opus-driven code changes are required.