Add AGENTS.md, CODEOWNERS, and PR template

[miguelcalderon]

What

Adds agent-first infrastructure and improves e2e test reliability.

Agent-first readiness

• AGENTS.md — Comprehensive guide documenting repo structure, example anatomy, all commands, how to add new examples, e2e testing workflow, CI setup, and code style rules. Designed for both AI agents and human contributors.
• CODEOWNERS — Assigns Web SDK team (nickel) as default reviewers
• PR template — Standardized checklist (type, formatting, e2e test, README, SDK version)

E2E improvements

• Added start:e2e scripts to javascript-vite and typescript-vite (were missing, now testable in CI)
• Fixed e2e-tests.sh to continue on failure and report all broken examples at the end, instead of silently masking failures
• Excluded svelte-kit and vue-composition-api from e2e (require pnpm, which CI does not have)
• Excluded svelte from e2e (uses sirv which isn't installed in CI)
• Excluded react from e2e (react-scripts/CRA crashes on Node.js 24)

Context

Part of the agent-first repos initiative discussed in #engineering.

Refs: https://nutrient.atlassian.net/browse/AI-450
Closes https://nutrient.atlassian.net/browse/WEB-4153

[vladimir-tikhonov-nutrient]

Nice improvement adding the missing start:e2e scripts. I pulled the branch locally and verified that the four touched examples pass the shared Playwright smoke test.

One thing I think is still worth fixing in the same area: scripts/e2e-tests.sh keeps iterating even if one example fails, and the script currently returns the status of the last command that ran. In practice that means a broken example in the middle can still be masked by a later passing example.

A tiny repro of the control flow is:

for cmd in false true; do ($cmd); done
echo $?
# 0

Would you mind making the runner fail fast with set -e (or collecting failures and exiting non-zero at the end)? Otherwise we can still end up with hidden failures even though the missing start:e2e cases are now covered.

[vladimir-tikhonov-nutrient]

I like having a checklist here. I think this specific checkbox will be noisy as written, though: several current examples don’t satisfy start + start:e2e literally.

The four Vite-based examples updated in this PR use dev rather than start, and the .NET examples don’t have a package.json at all. Could we scope this to new Node-based examples, or reword it to match the repo’s actual conventions?

[vladimir-tikhonov-nutrient]

Nice to see an agent guide landing here. One thing I’d tighten up before we rely on it: a few of these repo-wide rules don’t match the current examples yet.

For example, the Vite examples touched in this PR still expose dev + start:e2e, not start, several implementations use NutrientViewer.load() rather than PSPDFKit.load(), and asp-net / blazor don’t have a package.json at all.

Since this file is meant to steer both humans and agents, I think it would be safer to either describe the current state a bit more loosely (for example, “Node-based examples usually have ...”) or bring the repo into line with the guide.

[miguelcalderon]

Good catch! Fixed in 71a6675 — the runner now collects all failures and reports them at the end with a non-zero exit, so nothing gets masked. Went with the "collect and report" approach over set -e so we still get visibility into all broken examples in a single run.

[vladimir-tikhonov-nutrient]