Fix gyroscopic precession sign: roll term (row 0) had the wrong sign

[Zaffer]

The propeller gyroscopic torque had the wrong sign on the roll axis (row 0 of torque_inertia), in both dynamics (numeric) and symbolic_dynamics in crazyflow/dynamics/first_principles/dynamics.py.

The gyroscopic torque is -ω × h with net rotor angular momentum h = h_z·ẑ, which gives (-q·h_z, +p·h_z, 0) — the roll and pitch rows must have opposite signs (it is a cross product). The code had them with the same leading sign (-q·S, -p·S), so exactly one row is wrong.

The subtlety is which row. S = Σ mixing_matrix[-1]·ω_rotor uses the yaw-torque-direction row, which is the negative of the physical spin direction (the same row that correctly signs the drag reaction and the z spin-up/down term). So h_z = -prop_inertia·S, and substituting:

roll  (x) = -q·h_z = +prop_inertia·ang_vel[1]·S   # code had a leading "-": the bug
pitch (y) = +p·h_z = -prop_inertia·ang_vel[0]·S   # already correct

The z (reaction) row is already correct, and is what pins the convention: it is consistent only with spin = -mixing_matrix[-1]. So the fix is to flip the sign of the roll row, leaving the pitch row unchanged.

Verified numerically against -ω × h to machine precision on all axes; numeric == symbolic still holds for ang_vel_dot. Only matters for non-negligible prop_inertia / larger quads during yaw-type maneuvers; negligible for a Crazyflie.

Note: an earlier revision of this PR flipped the pitch row instead. That was incorrect — it left the roll bug in place and inverted the already-correct pitch term (roughly doubling the error). Now corrected to flip the roll row.