The following is given in the spirit of Paul Newman's famous line from Butch Cassidy and the Sundance Kid:
there are no rules in a knife fight!
Based on the question and comments, I think the OP would simply like to ged rid of the sinewave, to the maximum extent feasible, and also minimize the transient response to the 1 V step. So @Dan Boschen's advice about the Bessel LPF is good, but there is still the transient response and the overshoot: for a 5th order Bessel LPF, it is 0.76%. Re-using some LPF filter data from a paper I published in 1986, I have taken some liberties with the OP's stated values and obtained some results that may be thought-provoking, if nothing else. So consider the following model:
In the model, the signal source is a 20 Hz sinewave, with 0.1 V amplitude and riding atop a 1 V DC offset. The Butterworth and Bessel LPFs are third order and have 1 Hz noise bandwidths. The RC LPF has a time constant that is given by the output of a linear ramp: the starting value is 4 ms and the end value, reached after 0.5 s, is 0.25 s. So the RC LPF has a small time constant at the beginning, to quickly deal with the step transient, and then the noise bandwidth (which equals 1/4RC) is 1 Hz for the last 75% of the simulation. To further reduce the sinewave ripple, the RC LPF is followed by a simple running integrator that averages over one sinewave period, i.e., 50 ms, in this model. So it does a 50 point running average.
The next figure compares the three filters:
The traces are color-coded, as shown in the figure. Clearly the time-variant RC LPF did OK. The next figure is an expanded scale version, with only the Bessel and time-variant RC LPF responses:
I have not played around with the ramp values or tried a non-linear ramp, so I have no clue what might happen. But I think there is a point to me made: the more you know about the specifics of a given problem, and the more clearly you understand what you actually want to know or accomplish, the more opportunities you have in regard to solving the problem. By the way, the third order Bessel LPF has 0.75% overshoot, almost the same as the 5th order filter.
So now modify the first figure by deleting the RC LPF and ramp and clipper, so the input goes directly to the running integrator. The results are shown in the next two figures:
Of course, this will not work properly if the sinewave frequency is not constant. Anyway, this was all just intended to point out that sometimes it may be useful to think outside the box a bit.