Discriminate `anthropic` response based on type property

Anthropic returns a union of types based on the `type` property.

https://github.com/getsentry/sentry-python/pull/5715#discussion_r3027583313

The desired outcome is to access information required on AI Client Spans based on concrete response types based on the `type` property rather than probing the response object. Narrow down the type and then we can see based on the properties that exist on the response type which branches can be removed