Should be a setting that adds --enable-completion-model to the analysis server flags.

Changing the setting should automatically restart the analysis server, and we should skip adding the flag for SDKs versions that don't support this flag (to avoid a crash if you switch back to an old SDK).