fix(docs): retry flaky solc download in compile-solidity, fail-fast on persistent error

[AztecBot]

Summary

CI on merge-train/barretenberg (PR #22893, run https://github.com/AztecProtocol/aztec-packages/actions/runs/25226353436) failed because binaries.soliditylang.org DNS lookup intermittently failed while forge build was downloading solc 0.8.27 for the first solidity subdir (aave_bridge). The compile-solidity loop in docs/examples/bootstrap.sh then silently swallowed the failure (it runs under set +e), continued with the next subdirs (which now had solc cached and succeeded), and reported success. The TypeScript validation step then blew up on missing aave_bridge/AavePortal.sol/AavePortal.json etc., taking down the docs job.

This is unrelated to the actual change in #22893 (which only touches barretenberg C++).

Fix

In docs/examples/bootstrap.sh::compile-solidity:

• Retry each per-subdir forge build up to 3 times with 5s backoff to absorb transient solc-download failures.
• After the third failure, exit 1 from the subshell and propagate up via || return 1 so run_step's outer retry kicks in (and, if that also fails, the Slack notifier and the build correctly fail).

Test plan

• Cleared docs/target/solidity and ran ./docs/examples/bootstrap.sh compile-solidity locally; all three subdirs compile and the four aave_bridge JSON files referenced from docs/examples/ts/aave_bridge/index.ts are produced.
• CI on this PR exercises the docs job end-to-end.

Detailed analysis: https://gist.github.com/AztecBot/cab76a55438b3c9dc156829e8209197f

ClaudeBox log: https://claudebox.work/s/ea65e649c8b4b51f?run=1

[AztecBot]

Automatically closing this stale claudebox draft PR (no updates for 5+ days). Re-open if still needed.