#11882: fix: accept openai-codex/gpt-5.3-codex model refs

Fixes embedded Slack agents failing with "Unknown model: openai-codex/gpt-5.3-codex" by normalizing provider/model formatting before lookup and allowing gpt-5.3-codex version-suffixed variants (e.g. gpt-5.3-codex-YYYY-MM-DD).\n\nAdds unit tests for normalization + version variants. <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR updates `resolveModel` to normalize provider IDs and trim model IDs before looking them up in the discovered model registry, and extends the existing OpenAI Codex forward-compat fallback to accept `gpt-5.3-codex` version-suffixed variants (e.g. `gpt-5.3-codex-YYYY-MM-DD`). It also adds unit tests covering provider/model formatting normalization and the new version-variant fallback behavior. These changes sit in the pi-embedded runner’s model resolution path (`src/agents/pi-embedded-runner/model.ts`), which bridges configured providers/models (`cfg.models.providers`) with the model catalog discovered from auth/storage. <h3>Confidence Score: 4/5</h3> - This PR is close to safe to merge, but there is a provider-id mismatch edge case in the fallback path that can produce inconsistent provider identifiers. - Core normalization + new codex version-variant fallback look correct and are covered by tests; the main remaining concern is the fallback model using a normalized provider string even when it sourced config from a non-normalized provider key, which can break downstream provider-keyed behavior in that specific configuration. - src/agents/pi-embedded-runner/model.ts <!-- greptile_other_comments_section --> <!-- /greptile_comment -->