Does this tool's crossfade use a linear or a constant-power curve?
It uses equal-power: sin(t·π/2) for the fade-in and cos(t·π/2) for the fade-out. At the middle of the transition both gains equal 0.707, not 0.5, which avoids the volume dip a simple linear ramp produces at that point.
Can I request a crossfade longer than the shorter of two neighbouring tracks?
You can request it, but the real overlap is capped at the minimum between the requested value and the duration of each of the two neighbouring tracks. A 1.5-second effect between two 10-second blocks, with 3 seconds of crossfade requested, produces a real overlap of only 1.5 seconds on both sides.
Do two identical tracks get louder in the middle of the crossfade?
Momentarily yes, the sum can reach about 1.41 times the amplitude of either track alone at that exact point, because both equal-power gains equal 0.707 at once. A per-sample clamp in the final encoder keeps that peak from exceeding the PCM's ±1.0 range.