$n_1$ is a product of two relatively small 128-bit $p$ and $q$ so its easily factorable by a simple database look-up
Hint: use an approximation of the prime-counting function to comfort or challenge that reasoning.
A small public exponent attack should work for $n_3$, $n_4$, and $n_5$
Hint: $e=3$ is questionable if some conditions are met, see this. You haven't identified one of these conditions, nor why it would be met.
Is my hunch (that Fermat factorization will reveal a non-trivial factorization of $n_2$) correct?
You've got a non-quantitative justification of that hunch, which correctly identifies the non-trivial factorization you can hope to get. To decide if that works, quantitatively estimate what's the expected number of steps in the Fermat method, or just try it.
How can I go forward from here?
Hints valid for any RSA-related CTF: Write relations between unknowns as formulas. Estimate the expected bit size of unknowns. Try to identify a path towards a solution, as the sequence of unknowns you'll compute. Put to use the fact that $n_3$, $n_4$, $n_5$ are prime. Basic tools are algebra, the FLT and the CRT (plus, in some advanced cases, Copersmith's theorem). Show us your progress in the question if you hope for more hints.
Does the suspicious low public exponent encryption of $n_3$, $n_4$, and $n_5$ have anything to do with factoring $n_2$?
You don't present any argument towards this. And $e=3$ is not used in conjunction with $n_2$. On top of it, breaking poorly devised RSA encryption with whatever odd public exponent leads to factoring the modulus as a byproduct only when it's recovered a private exponent, or exploited the implementation, or the plaintext somewhat helps the factorization.