I have a sample of 150 first day (IPO) stock returns. Due to
"supposed" underwriter support I feel the sample is biased and I want
to censor all the negative and zero values to zero, using only the
positive values to estimate a "true mean" for the entire sample, which
I suspect to be lower than the actual mean I obtained. this method
assumes normal distribution, which is fine.

I am looking for a methodology (preferably in excel) that will allow
me to calculate this "true mean" for my sample.

Why do you suspect to have "tru mean" lower than the actual mean. If
you take out all the negative and zero values out of your sample data,
the new mean must be higher than the old mean, since negative and zero
values were dragging your mean down.
Do you have data imported to Excel?