The direct trajectory is an ellipse intersecting the pole:
We do not have the semi-major axis of this trajectory, however, so we must obtain that first.
For that, we can use one of my favourites, the equation for the orbital radius in terms of the periapsis $P$, apoapsis $A$ and the distance along the apsis line $d$
$$r = P + d\frac{A - P}{A +P}$$
For our case, $d = P$, so we have:
$$r = P + P\frac{A - P}{A +P}$$
Solving for the quantity we want, we have:
$$P = \frac{Ar}{2A - r} = 2070 km$$
We then have all the parameters we need to use the vis-viva equation
$$v = \sqrt{\mu\left(\frac{2}{r} - \frac{2}{A + P}\right)}$$
We want the velocity when at Phobos, the apoapsis of that trajectory, so
$$v_A = \sqrt{\mu\left(\frac{2}{A} - \frac{2}{A + P}\right)} = 1290 m/s$$
Now, we must not forget that this is a problem in 3 dimensions:
Which means the apoapsis velocity of $1290 ms/$ and the orbital velocity of Phobos are orthogonal:
The delta-v cost of the burn is therefore 2500 m/s, and is 30 degrees zenith from retrograde.
Cheaper and slower solution:
Launch prograde instead, up to almost escape velocity from Mars (delta-v cost: 886 m/s, $\sqrt{2} - 1$ times the orbital velocity of Phobos).
Then, at great distance, where the orbital velocity is just a few m/s, change the inclination and periapsis at almost no cost and fall back towards the Martian poles with a more violent impact.