non-docker + non-batch mode isn't using the correct defaults

At least for gpt-oss models if you haven't specified reasoning level it the non-batch mode for non-docker script will default to using reasoning High but truncate tokens to max of 100. the non-batch mode should be aligned with the batch mode in terms of defaults and behavior.

Compare the non-batch, non-docker execution path with the batch path, focusing first on defaults for gpt-oss models and the token limit. Done means both modes use the same reasoning-level default and do not unexpectedly truncate tokens to 100.

Written by the indexing model from the issue text.