Accept provider-native model identifiers in inference profiles

Instance admins can now save a Fireworks or OpenAI inference profile using the model identifier the upstream provider actually expects, for example accounts/fireworks/models/minimax-m3 or gpt-4o-mini, without it being rejected by the profile form. Existing profiles that store the older provider-prefixed value continue to route correctly, so no migration is needed. The dashboard placeholders, development seed data, and self-hosting guide were updated to show the correct provider/model examples.