1) It varies by guidance type, and by geometry conditions.
For example, all beamriders use pure-pursuit, because they have no concept of the engagement geometry, they just head to where they are pointed. Most guidance types, incl. semi-active and datalink / command-guidance, use some variation of collision course and/or intercept geometry. This is very clear in this example: https://i.imgur.com/en30qqB.png .
Plot-to-intercept is not however the cure-all. There are transient cases, where pure-pursuit is actually a better choice, because under certain conditions the intercept-point logics can be exploited by the adversary. (If you are familiar with DCS World, the "snake right and left once they launch BVR missiles at you, to drag them down" is an example of this. It's a gamey tactic that exploits the "intercept course uber alles" logic and AFAIK has no basis on RL evasion tactics.). So in some cases you have switchover between collision/interceot and pure-pursuit, and back again, as the geometry changes.
2) Hard to say without seeing it ourselves in a save. You can attach a file or post an external link once you reach 10 posts. Be aware that we are "slightly busy" these days so it may take longer than usual to investigate.
3) Please explain "TMA-based algorithm" (I assume it does not stand for "Target Motion Analysis").