OnTriggerEnter Or OnCollisionEnter?

One is not "better" than the other; they do different things. OnCollisionEnter is called when two colliders hit. OnTriggerEnter is called when a collider enters a trigger (i.e. a collider with "IsTrigger" set).

True "collisions", used in conjunction with non-kinematic rigidbodies, are for when you want to simulate the physical effects of two objects colliding - the conservation of momentum, forces applied based on the angle of incidence etc. OnCollisionEnter is called when these objects first hit each other and gives you properties of the point of collision, but the physics engine itself automatically resolves changes in motion of the two bodies etc.

OnTriggerEnter is used (with kinematic or non-kinematic rigidbodies) to detect when one object has entered another's trigger. Note that, in this case, the physics engine does not prevent the object from doing so - unlike a collider, you do not "bounce off" a trigger. It can be thought of as an invisible "zone", and you can decide what happens to the object having entered that zone.

Either can be used for a fighting game - it really depends how you are planning to control the kinematics of the characters.

Neither. You want to use Raycast/Linecast/Boxcast etc to actually tell if in attack hits generally.