fix(docs-examples): retry yarn add and surface errors on final attempt

[AztecBot]

Summary

PR #23344 (merge-train/spartan) was dequeued from the merge queue on the x8-full grind with docs/examples/bootstrap.sh execute exiting rc=1 and no diagnostic in the log (ci.aztec-labs.com/480b0d23d93a3e87):

09:57:13 Setting up aztecjs_advanced...
09:57:49 + rc=1
09:57:49 + exit 1

setup_project() in docs/examples/ts/aztecjs_runner/run.sh runs each yarn add with > /dev/null 2>&1. Under set -euo pipefail, a transient yarn add failure (link or registry flake) silently exits the runner with no output. aztecjs_advanced is the largest link install in the suite (6 @aztec/* packages), so it's the one that catches a bad day.

Change

Add a small yarn_add_with_retry helper:

• Retries yarn add up to 3 times, sleeping 5s between attempts.
• Stays silent on success and on early retries (preserves current "happy path" log volume).
• Unsilences stderr on the final attempt so the next flake surfaces in CI logs and is debuggable.

All three yarn add call sites in setup_project are routed through it (link deps, npm deps, and typescript/tsx).

Test plan

• Local sandbox: docs/examples/ts/aztecjs_runner/run.sh advanced succeeds end-to-end.
• CI: docs/examples/bootstrap.sh execute passes on a clean grind; on a flake, the failure shows yarn add output instead of an empty rc=1.

---

Created by claudebox · group: slackbot