I'll start by saying, we recommend afni_proc.py, which is a superscript for accomplishing most of your fMRI analysis needs. You can see more info in the docs and class handouts.

To your specific questions on the fairly abstract description of your task:

1. Do not do #1 without some hefty processing (e.g. converting to signal change, detrending, censoring, many more to list)
2. This is a potentially useful way to go
3. Unless you have a specific "wouldn't this be great for methods" question, I wouldn't advise on #3. The fundamentals of measuring "activation" in a block design using a "let's average some runs" mentality are often better addressed using regression.
4. My main recommendation is to use afni_proc.py to process the two runs at the same time. This will take care of the concerns in #1, give you the power to do #2, and essentially do #3. Using afni_proc.py, you could also setup GLT "contrasts" to compare run 1 and run 2 if that's of interest.