Oh right. So you're kinda picking an actual number to restrict delta to be less than first, before generalising it for all epsilon?
Well essentially the motivation behind making delta to be less than 2 is to deal with the |x+1| in the |x+1||x-1| stage.
When we're at |x+1||x-1|, we want to make this less than eps.
We can control |x-1| directly, since it's less than delta, so we can make it less than eps/4 say (we could also do a lot of other things, like eps/6, or whatever; the choice for delta is non-unique).
If we could just force |x+1| to be less than 4, we'd be done, since then we'd have |x+1||x-1| be less than 4*eps/4 = eps.
The idea is that when we're "near" 1 (since x is approaching 1), |x+1| is going to be bounded; it can't get too big.
Precisely speaking, if we restrict our consideration to be within 2 units of 1, then |x+1| will always be less than 4.
So by taking delta to be smaller than* 2 AND smaller than eps/4, we get |x+1| < 4 AND |x-1| < eps/4.
The way to make these simultaneous "smaller thans" for delta is to make it the min of those two things, and then let |x-1| be less than delta.
*
delta actually doesn't need to be smaller than them, just equal to them. Then by letting |x-1| be less than this delta, we get |x+1| < 4 and |x-1| < eps/4, which is what we wanted.