fix: trim surrounding whitespace in <title> same as <desc>

[si-kui-a]

Fixes #2052.

Problem

textElems in plugins/_collections.js explicitly included 'title' alongside 'pre' and the genuinely-rendered text-content elements (text, tspan, etc.) — the elements where surrounding whitespace is semantically meaningful and must be preserved verbatim. <title> is a non-rendered accessibility/metadata element — like its sibling <desc> and <metadata> in elemsGroups.descriptive — where whitespace has no such significance, but it was the only member of that group also present in textElems.

lib/parser.js's sax.ontext handler branches on textElems membership: members keep raw (untrimmed) text, non-members get text.trim(). Since title was in textElems, <title> My Title </title> kept its leading/trailing whitespace while the structurally identical <desc> My Desc </desc> got trimmed to "My Desc" — the exact asymmetry reported in #2052.

Fix

Remove 'title' from the textElems Set literal (one line). Confirmed via repo-wide code search that only lib/parser.js and lib/stringifier.js reference textElems besides the definition itself, so the change is fully self-contained.

lib/stringifier.js also branches on the same textElems set to decide whether an element's children get pretty-mode indentation/newlines, so this also makes <title> format consistently with <desc> when js2svg pretty output is used — same underlying membership change, not a separate behavior. This surfaced as one existing fixture test failure (inlineStyles.16), whose expected output I've updated to match: <title>button</title> now gets the same indented multi-line layout <desc> already receives in pretty mode.

Testing

Added a regression test in lib/parser.test.js reproducing the exact before/after asymmetry from #2052 (title now trims identically to desc).

Ran the actual suite locally (not just hand-traced):

• pnpm install --frozen-lockfile
• pnpm vitest run — 521 passed, 3 skipped. Confirmed the new test genuinely catches the bug by temporarily reverting the fix and re-running (it fails as expected), then restored the fix and re-ran (passes).
• pnpm lint (eslint + prettier) — clean.
• pnpm typecheck (tsc) — clean.

Did not run test:bundles / test:regression (the full qa script's remaining steps) — those build the rollup bundle and pixel-diff a large external SVG corpus, which felt disproportionate to verify for a one-line Set membership change already covered by the unit/fixture suite above; happy to run them too if maintainers want that before merge.

🤖 Generated with Claude Code

https://claude.ai/code/session_01HG5wiw3Z8QdjNWb5bFDLiA