Bedrock model openai.gpt-oss-120b invalid/inaccessible in us-east-1, breaks chat e2e test

## Summary

During deployment/e2e testing of PR #60 (fixing the CUSTOM_JWT/SigV4 auth mismatch), `e2e/chat.spec.ts`'s "agent returns a response" test still fails after the auth fix, this time with an error surfaced directly from Bedrock:

```
ValidationException: The provided model identifier is invalid.
Bedrock region: us-east-1, Model id: openai.gpt-oss-120b
```

## Where it comes from

`agent/default/app/MyHarness/harness.json` sets:

```json
"modelId": "openai.gpt-oss-120b"
```

This value hasn't changed since PR #30 (well before PR #60), so it's a pre-existing issue unrelated to the auth-mismatch fix — it just happened to be the next failure encountered once auth was working.

## Likely cause / next steps

- Check Bedrock model access for `openai.gpt-oss-120b` in `us-east-1` for this account — the model may not be enabled/subscribed.
- Alternatively, update the harness's `modelId` in `agentcore.json` (source of truth per `CLAUDE.md`) to a model that is available in the target region/account.
- After fixing, redeploy (`agentcore deploy`) and rerun `pnpm test:e2e e2e/chat.spec.ts` to confirm the "agent returns a response" test passes.

## Context

Found while running e2e tests against PR #60: https://github.com/waltmayf/agentcore-amplify-fullstack/pull/60#issuecomment-4904912770

---
*Imported from `waltmayf/agentcore-amplify-fullstack`#63 — originally filed by @app/github-actions on 2026-07-07. Migrated to open-source repo; cross-references updated post-import.*

Start with agent/default/app/MyHarness/harness.json and verify the modelId against the source-of-truth guidance in CLAUDE.md and agentcore.json. Check whether openai.gpt-oss-120b is available in Bedrock for us-east-1, then redeploy with agentcore deploy and run pnpm test:e2e e2e/chat.spec.ts. Done means the agent returns a response without the model identifier error.

Written by the indexing model from the issue text.