Example reference vs local copy reconciliation

[rob-metalinkage]

Keeping a stable accessible local copy to ensure bblock processing robustness is valuable - we should have a "fetch-ref" option that caches the example - and we can add a "sync check" option to check ref is available and if it has changed.

[avillar]

What do you think of this? Add a per-ref cache boolean flag (in examples.yaml snippet refs and tests.yaml refs), defaulting to off. When set:

1. On every run, we still attempt a live fetch of the ref.
2. On success, the fetched content is written to a persistent cache directory (keyed by URL hash, reusing/extending the existing remote_cache mechanism, but stored somewhere --clean won't wipe).
3. On fetch failure (network error, timeout, HTTP error status, etc.), we fall back to the cached copy if one exists, and log a warning identifying the ref and the failure reason.
4. The build only fails if there's no live fetch success and no cached copy available.

This intentionally treats any fetch failure the same way, since a persistently failing or unreachable remote is exactly the case caching is meant to protect against. Nothing is ever swallowed silently: whenever the cache is used as a fallback, a warning is logged so it stays visible in build output and reports.

[rob-metalinkage]

hmm - i think it should compare and if different throw a validation failure unless the cached version is flagged as the point of truth.

if flagged as the point of truth, then only fetch on a flag to force?