Noise "breathing" effects are a well known problem with recorders of this type and they are a PITA
An attenuator will reduce the risk of overloading the mic input but make other matters worse as the mic gain will be wound up higher and these circuits are designed down to a price not for any "low noise" attributes. They often pick up noise from the motor circuitry as well and, as you say, the AGC will still be winding the qain up and down.
You would be better buying a quality recorder, however, there is a scheme which might work.
Build yourself a small two channel mixing circuit. Into one channel you feed your telephone pickup, into the other the output of a high frequency oscillator - start with a frequency you "just can't hear". The oscillator provides a constant level which will keep the AGC happy - you will have to experiment with it's level to get the best recording. Since the frequency is above hearing you won't hear it on playback - nor is it likely to get through the limited bandwidth of the tape system anyway.
If there is any sort of bandwidth limiting before the AGC monitor point, or it causes interference with the bias oscillator, you will have to drop the frequency lower and use a notch filter on playback.
The other mixer channel can be used to control your telephone signal level so that it does not overload the mic input.
Yes. The attenuator in this case would be a potentiometer. Try a 10K ohm skeleton preset "pot". Wire the source to the top (hot), and bottom (ground) terminals. Wire the lead to the mic jack to the centre and bottom terminals.