You are (or I should more accurately state that the math in the link you are presenting is) assuming Trump and Cruz are perfectly negatively correlated (that a voter who leaves Trump will go to Cruz and that a voter who leaves Cruz will go for Trump). Do you have any data to back that up? It is quite a big assumption in such a crowded field and doesn't feel correct to me. If not, you are neglecting a whole lot of other error terms in the proper propagation of error math. If you include all of the propagation of error math, and proper formulas, then you end up basically with the rule-of-thumb worst case scenario that I posted above (and anyways it is just safest to go with the worst case scenario than to bicker about tenths of a percent).
For example, suppose that we were to use percentages where that math you linked was correct (roughly 50% voter support) and that they truly were negatively correlated. Then p1=50%, p2=50%, n=602 and the math comes up with an error of:
1.96 * ((0.5 + 0.5 - (0.5 - 0.5)^2) / 601)^0.5 = 7.995%. Oh, wow, I said take 4% * 2 = 8%, I'm so far off from 7.995%. Also, like you stated, your own formula isn't valid for small percentages to begin with. So we can't use your linked math when the percentages go down to the 20% range. That math artificially gives a too-low MOE as you stray very much from 50% politician support. For example, try polling 2 people and see what happens if they both support Trump. Suddenly that math has a 0% margin of error, Trump will win the race with 100% of the vote and 0% MOE. Or, you can realize the math isn't applicable. The safest thing to do is just resort to rules of thumb, or do the full proper math that doesn't vary much from the rule-of-thumb anyways.
Rather than get into a battle of propagation of error and assumptions, since that will likely go nowhere, lets just go to a simple table:
http://www.washingtonmonthly.com/archives/individual/2004_08/004536.php
Trump was 5% ahead in your posted poll, with a 4% poll MOE. Thus, if the poll was done perfectly there is a 90% chance that Trump was in the lead in your poll. Since the poll is clearly not done perfectly (self-selection, bias either intentional or not, time differences, etc), there is a less than 90% chance that the poll showed that Trump was in the lead.