Nick,
1) seems to be reasonable to me: Don't worry about the temporal discontinuities because the correlation computation only focuses on the relationship at the same time point (multiplication), but not across time points (only summation).
2) is fine, but would probably render similar results as 1).
The problem with 4) is that you could only remove the *average* condition effect (across trials) from the data, but not the variability across trials.
Gang