Strengthen login attribution test to exact-match every provider

The exhaustive walk previously only asserted is_some() for each login
provider descriptor's id and display name. That would have passed even
if a provider's hint pointed at the WRONG provider - exactly the class
of bug that produced "OpenAI credentials are active" after an
Anthropic login.

Now compute the expected canonical hint from each descriptor's target
(mirroring auth_provider_hint_for_login_provider, the single source of
truth) and assert the exact value across the id, display name, and
every alias. Verified out-of-tree: 159 label-checks across all 46
descriptors resolve to the correct hint with zero collisions.