For comparison, here is the dataset for Royal v. Sarkozy 2007, without sample sizes.
(edit: here is a dataset for the first round 2012)
Hopefully, those datasets can be helpful to some people...
I took the data from Wikipedia, and there might be better datasets out there.
To play with the data you can start with the R code provided by Arthur Charpentier. Here are the graphs generated with the 2012 polls only, with smoothing by splines and 95% confidence intervals.