feat(AIC-3449): refactor AI Config reference into a dataclass

[aknight-ld]

Making the relationship between config and variation keys explicit for easier validation. No functional changes beyond this minor refactor.

---

Note

Overview
Breaking API change: EvaluationsModule.run() no longer accepts separate ai_config and variation string arguments. Callers must pass a single AIConfig(key=..., variation=...) frozen dataclass when seeding a run from an existing AI Config variation.

The new type is exported from launchdarkly_ai_server and launchdarkly_ai_server.evaluations. Validation now checks ai_config.key and ai_config.variation for blank values (replacing the old paired-parameter errors). Fetching, merging with generation, tools, judges, and logging behavior are unchanged aside from using the struct’s fields.

^{Reviewed by Cursor Bugbot for commit 2bb40bb. Bugbot is set up for automated code reviews on this repo. Configure here.}