Extrapolation from 5 Hz down to 0 Hz has always been problematic, and I can almost promise that changes in 2.0.110.1 are not the last.
Previous versions had small bug for several months; real and imaginary part of 0 Hz point was zeroed with all LF slopes, but zeroing was not (accidentally) done to 0 Hz point of "mirrored" frequency response which fills the second half of the FFT buffer. Now both ends are zeroed, but possible problem is that filter of woofer could extend down to 0 Hz. 0 Hz point is not necessary to zero if there's no HP slope at LF.
Much older versions detected LF slope and zeroed or copied values from the next frequency point to 0 Hz if slope is < 3 dB/oct. That was expected to be better approach, but somehow I wasn't happy with it. Maybe it should be evaluated again.
Full range flat frequency response produces natural DC offset to impulse response. That's not a problem, but some DC variation (easily visible in step response) could exist with almost any real life IR depending on combination of sample rate, FFT length, taps and window function because limited part of IR could include DC offset also with high passing slopes.
So there is no perfect solution for FIR with truncated frequency range and limited taps. Some errors will exist at LF.
