Since the theater commander runs once per campaign action, missions that
do not have aircraft available may be checked more than once a turn.
Without deduping requests this can lead to cases where the AI buys
dozens of tankers on turn 0.
Fixes https://github.com/dcs-liberation/dcs_liberation/issues/1470
The doctrine/task limits were capturing a reasonable average for the
era, but it did a bad job for cases like the Harrier vs the Hornet,
which perform similar missions but have drastically different max
ranges. It also forced us into limiting CAS missions (even those flown
by long range aircraft like the A-10) to 50nm since helicopters could
commonly be fragged to them.
This should allow us to design campaigns without needing airfields to be
a max of ~50-100nm apart.
This makes it so that the mission planning effects are applied only if
the package can be fulfilled. For example, breakthrough will be used
only if all the BAI missions were fulfilled, not if they will *attempt*
to be fulfilled.
Consider BAI missions planned this turn when determining if a control
point is still garrisioned for preventing breakthrough.
This isn't very accurate yet since the HTN isn't checking for aircraft
fulfillment yet, so it might *plan* a mission to kill the garrison, but
there's no way to know if it will be fulfilled.
This improves the AI behavior by choosing the stances non-randomly:
* Breakthrough will be used if the base is expected to be capturable and
the coalition outnumbers the enemy by 20%.
* Elimination will be used if the coalition has at least as many units
as the enemy.
* Defensive will be used if the coalition has at least half as many
units as the enemy.
* Retreat will be used if the coalition is significantly outnumbers.
This also exposes the option to the player.
IADS that are in detection range (but not attack range) of missions will
be targeted at very low priority. These will typically only be planned
when no other targets are in range.
The players can still manually assign strike missions on other target
types since that's sometimes better for player waypoint generation (one
waypoint per unit is nice for SAMs), but it's bad for the AI so by
default we should exclude non-buildings.
This also prevents double targeting of groups, since they might have
been identified by other missions as well.
We already did some of this, but since we were excluding specific TGO
types rather than only allowing building TGOs we were often missing
things (missile sites, coastal defenses, and EWRs, it seems).
Garrison groups should be preferred with the following priority:
1. Groups blocking base capture
2. Groups at bases connected to an active front line
3. Rear guard units
Previously they were being prioritized based on the distance to the
closest friendy control point, which is similar to this but an
aggressively placed carrier could throw it off.
This alters the DEAD task planning to be the *least* preferred task, but
prevents other tasks from being planned unless they are excepted to be
clear of air defenses first. Even so, missions are a guaranteed success
so those other missions will still get SEAD escorts if there's potential
for a SAM in the area.
This means that air defenses that are not protecting a more useful
target (like a convoy, armor column, building, etc) will no longer be
considered by the mission planner. This isn't *quite* right since we
currently only check the target area for air defenses rather than the
entire flight plan, so there's a chance that we ignore IADS that have
threatened ingress points (though that's mostly solved by the flight
plan layout).
This also is still slightly limited because it's not checking for
aircraft availability at this stage yet, so we may aggressively plan
missions that we should be skipping unless we can guarantee that the
DEAD mission was planned. However, that's not new behavior.
An HTN (https://en.wikipedia.org/wiki/Hierarchical_task_network) is
similar to a decision tree, but it is able to reset to an earlier stage
if a subtask fails and tasks are able to account for the changes in
world state caused by earlier tasks.
Currently this just uses exactly the same strategy as before so we can
prove the system, but it should make it simpler to improve on task
planning.
