If its RNAseq - if the batch/sample mapping is known, we include it in the design for differential expression, or if not doing differential expression, we either apply limma::voom or DESeq2::rLog and then use limma::removeBatchEffects().

Depends what you are doing. For differential gene expression, you can just include batch in the design, and that's enough. For other applications, you might try ComBat or RUVSeq

If you don't know batches, may be RUVSeq helps but it depends on non-differential gene expression to estimate surrogate variables. PCA regression performs well. Its included not in sva package as sva_network function. Plot the PCs and see how many PCs to regress out. sva also has function to estimate the no. of PCs to regress out (num.sv) but from my experience it often overestimates number of PCs.