I always thought that the parameters for insert size estimation are misguided and it is never clear what happens when one makes use of them. The concept of 'concordant pair' or 'mapped in the proper pair' imposes a certain view of the genome should be like but that may not be at all what the current genome is. It is from an era where people thought that non-concordant must mean an error.

I would use a mapper that does not have any expectation of the "right" mate sizes are and work backwards see how the mates map individually and how far they are from each other. Could be even mapped in single end mode.

I normally use GSNAP to map the mate pairs back to the closest reference genome (with stringent settings: no soft clipping, low number of mis-matches etc) and then plot the mapping distance between the read pairs. The settings --pairmax-dna and --pairdev allows you to specify the insert size and standard deviation to be used for calling concordant/discordant. I hope this helps!

BTW, GSNAP will be exceptionally slower compared to other aligners/mappers unless you split them up!