fix: make result extrema scale safe

[petergstfsn]

Summary

• add a shared finiteExtrema reduction for ordinary arrays, sparse arrays, typed arrays, and projected values without input copies or argument spreading
• migrate unbounded result-field extrema in Core result assembly, solver diagnostics, static envelopes, imported/derived fields, transient ranges, and result UI paths
• add supported-scale regression coverage for displacement, stress, reaction, imported fields, element-sized arrays, empty fields, and nonfinite values
• make BROWSER_SOLVE_LIMITS the explicit product-limit authority and distinguish the approximately 100k validation benchmark from the guarded 150k-DOF ceiling
• make STEP meshing's quadratic fallback derive its node guard from the authoritative browser limit instead of a stale hardcoded 100k-DOF value

Public contracts and review flags

• @opencae/core now exports finiteExtrema and FiniteExtrema.
• The helper ignores NaN and both infinities, returns null for empty/all-nonfinite inputs, and treats ordinary/sparse/typed arrays consistently. Existing result-field fallbacks retain a finite 0..0 range for empty fields.
• The browser product ceiling remains 150,000 DOF; this PR does not raise it. The mesh-intake preflight now observes that existing authority instead of imposing a separate 100,000-DOF limit.
• Result legend midpoint labels now use the finite range midpoint instead of allocating and sorting a full result array for its median. Units and label/result shapes are unchanged.
• @opencae/mesh-intake gains one existing workspace dependency on the lightweight @opencae/solve-pipeline/limits entry point. No third-party dependency was added.

Verification

• pnpm install --frozen-lockfile — passed
• git diff --check — passed; the repository has no formatter command or root lint job
• pnpm --filter "./packages/*" --if-present lint — passed (the five package lint scripts are typecheck aliases)
• documented package builds — passed
• pnpm --filter "./packages/*" --if-present test — passed: 259 tests
• node scripts/verify-cloudflare-config.mjs — passed
• pnpm typecheck — passed
• focused scale/result suite — passed: 160 tests
• pnpm test — passed: 153 files, 1,590 tests
• pnpm verify:step-selection-browser — passed in real Chrome for 1 mm, 25 mm, and 500 mm fixtures; detailed/balanced face IDs and fingerprints matched with no console or page errors
• pnpm build:cloudflare — passed
• pnpm --filter @opencae/web check:bundle — passed: initial JS 63,822/179,200 bytes gzip; total JS 1,343,404/1,536,000; largest lazy gzip 141,618/307,200; largest lazy raw 445,787/512,000; precache 24,580,590/27,262,976
• pnpm verify:perf — passed its production build and local HTTP/startup checks (402 ms startup); its optional Playwright browser probe was not configured and was skipped
• node scripts/verify-100k-solve.mjs — attempted with host Chrome access, but timed out waiting for the Chrome solve-benchmark result after approximately nine minutes; the serial WebKit leg did not start. This run supplies no new cross-browser or 150k performance certification.

Assumptions

• BROWSER_SOLVE_LIMITS is authoritative because current production preflight and worker orchestration import it, and no contrary code authority was found.
• Small fixed-size vectors and bounded collections remain on spread-based reductions where the argument-count failure cannot occur and the expression remains clearer.
• Nonfinite result samples remain present for validation/diagnostics; only extrema calculation ignores them.

No deployment, production service, migration, secret, or infrastructure operation was performed.

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Updated (UTC)
✅ Deployment successful! View logs	opencae	eb0b12d	Aug 28 2026, 07:27 PM