fix: repair all failing tests across utility modules

[stooit]

Summary

Fixes all 16 failing tests (now 60/60 pass) across the utility library, implementing missing functionality and repairing edge-case bugs. No test files modified, no dependencies added. tsc --noEmit is clean.

Changes by module

• calculator.ts — divide now throws on a zero divisor (b === 0, also catches -0) instead of returning Infinity/NaN.
• string-utils.ts — wordCount uses trim().split(/\s+/) so consecutive whitespace no longer produces empty entries. truncate implemented: cuts at a word boundary with "..." counting toward maxLength, returns unchanged when within limit, never exceeds maxLength.
• task-manager.ts — implemented remove (boolean), update (returns false for unknown id; only applies provided keys), and sortBy (priority high→medium→low, createdAt oldest first). sortBy throws on unknown field with a never exhaustiveness guard.
• date-utils.ts — fixed off-by-one in formatRelative: Math.round(abs(diffHours)/24) so 36h maps to "2 days ago"; future dates stay symmetric.
• validator.ts — isEmail accepts arbitrary subdomain depth and long TLDs (e.g. .museum); isUrl accepts URLs with ports (e.g. http://localhost:3000) while still rejecting non-http(s) protocols.

Verification

• bun test → 60 pass, 0 fail
• bunx tsc --noEmit → clean

Reviewer notes / assumptions

• divide is a semver-major behaviour change for consumers relying on Infinity/NaN propagation. Required by the test suite; flagged for changelog.
• update's !== undefined guard means description cannot be cleared via undefined — kept as the intentional convention.
• isEmail/isUrl are format checks, not trust-boundary controls. localhost/internal IPs accepted by design (unchanged from prior behaviour); add SSRF protection separately if a URL feeds a server-side fetch.

🤖 Generated with QuantCode