When switching to a new search backend, validation could fail due to the

search backends attempting to access the "default" connection, which is

the forwarding backend, and then failing to find the right engine

information.
To address this, we now register a new, empty connection alias. We set

this to the new engine and configuration, and then validate using that.

This gives the backend a stable connection to hold onto.
This makes use of modern Haystack's ability to modify and reload a

registered connection, which is the original purpose of the forwarding

backend. This is the reason we're not just using the forwarding backend

for this purpose. However, the forwarding backend is also more complex,

loading state from the site configuration and handling cache-based

syncing, with fallbacks on errors. We don't need or want all that, so we

just work with what Haystack provides us.

Tested switching between backends, with successful and failed validation

attempts. I no longer had any issues. I also added temporary logging to

Haystack to verify configuration and connection states.