fix: repair all failing tests in utility library

[stooit]

Summary

Fixes all 16 failing tests across 5 source files. Test suite now 60 pass / 0 fail (was 44/16), bunx tsc --noEmit clean. No test files modified, no dependencies added.

Changes

• src/calculator.ts — divide throws Error("Division by zero") when the divisor is 0 instead of returning Infinity.
• src/string-utils.ts — wordCount splits on \s+ after trim (fixes phantom words from consecutive spaces); truncate implemented with word-boundary cut-back and ellipsis counted toward maxLength.
• src/task-manager.ts — implemented remove (returns Map.delete boolean), update (applies only defined fields, false for unknown id), and sortBy (priority high→medium→low, status, createdAt oldest-first).
• src/date-utils.ts — formatRelative day bucket uses Math.round(Math.abs(diffHours)/24), fixing the off-by-one (36h → "2 days ago") symmetrically for past and future.
• src/validator.ts — isEmail accepts subdomains and TLDs of 2+ letters (user@example.museum); isUrl accepts URLs with ports (http://localhost:3000).

Verification

• bun test → 60 pass, 0 fail
• bunx tsc --noEmit → exit 0
• Independent review pass: fixes confirmed as genuine implementations, not assertion-shaped tricks; regex changes are linear (no ReDoS); no regressions on previously-passing behaviour.

Assumptions & notes

• Scope limited to what the tests require, per task constraints. Two non-blocking design notes surfaced in review and intentionally left as-is (untested, within spec): update cannot clear an optional field via undefined (JSON-merge semantics); truncate's word-boundary heuristic can under-fill the budget when the first word is long. sortBy status ordering (in_progress > pending > completed) is not test-specified.