When a plain text file is opened (no matter if "Text" or "Text - choose encoding" was used), the filter settings (charset, Line End, is BOM present: see SwAsciiOptions) are forgot, and last used for export are shown if "Edit filter settings" is requested. In case, say, I open a UTF-8 with BOM, I'll get a Windows-1251 Cyrillic file upon save.
The settings should persist with the open file, to not require user to re-enter them on save (a user might have no idea what they were if auto-detection was used).
You are right the same type should be used for save.
And by default for "save as". But here the user should have the option to change settings.
As has been remarked elsewhere -- for example 82254, "FILESAVE: UTF-8 BOM removed from CSV when saving file" -- this is how LO should act for all documents it can read and save in plain text form. For predictability, consistency, accuracy, and usability.