In the Microarray if the data is not center around 0 on MA-plots, it is normalized with LOESS. I was not able to find if edgeR performs something like loess normalization for log2Fold changes as it seems that the values are always centered around zero.

The centering of estimated log fold changes around zero is a consequence of the estimation of the size factors (as they are called in DESeq2) or normalisation factors (edgeR).

The kind of non-linear operations on microarray data as entailed by loess-fitting in the MA-plot are motivated by non-linearities in the data, like scanner saturation and background fluorescence. These effects do not appear in decent RNA-Seq data and the authors of edgeR and DESeq2 have not seen a need for introducing such non-linear fudges.