Fix invulnerability damage and armour

[okx-code]

If a plugin changes the damage in EntityDamageEvent, it can cause the INVULNERABILITY_REDUCTION damage modifier to be set to 0 and render invulnerability damage reduction, added in #11599, ineffective.

This happens because a float (the lastHurt value in a Player) is being compared to a double (stored in the modifiers of EntityDamageEvent), and a float can be greater than a double, even if they are the same noiminal value:

jshell> 3.2 < (float) 3.2
$2 ==> true

Due to lossy floating point precision, 3.2 becomes 3.200000047683716, which indeed would be greater than 3.2.

float lastHurt is the result of casting the double from EntityDamageEvent#getFinalDamage, which causes this imprecision. The fix is to cast the doubles from EntityDamageEvent to floats when comparing them here. This ensures that if they are the same value, then one will not be less than the other, because they have both been cast from the same double value. The opposite approach, casting a double to a float and back to a double again, does not work, because is lossy. We cannot really convert lastHurt to a double without touching a lot of the damage processing code.

I have also taken the liberty of fixing how armour damage is calculated. In the #11599 PR, armour would still take damage even if the damage should have been blocked due to invulnerability. This change will now take that into account. It's also necessary to individual cast each component to a float, rather than adding the doubles like before, otherwise the armourDamage may be marginally greater than 0. I've done this in a similar way to the LivingEntity#computeAmountFromEntityDamageEvent function.

The BASE damage and INVULNERABILITY_DAMAGE modifers may be slightly different doubles, but represent the same float. The INVULNERABILITY_DAMAGE represents the BASE modifier, but casted to a float and back to a double. Therefore we must cast BASE and INVULNERABILITY_DAMAGE to floats before adding them.

To test this PR, you can use the following code and run the /test command (plugin.yml not included)

public class TestPlugin extends JavaPlugin implements Listener, CommandExecutor {

    @Override
    public void onEnable() {
        getCommand("test").setExecutor(this);
        getServer().getPluginManager().registerEvents(this, this);
    }

    @Override
    public boolean onCommand(CommandSender sender, Command command, String label, String[] args) {
        ServerPlayer handle = ((CraftPlayer) sender).getHandle();
        handle.hurtServer(handle.serverLevel(), handle.level().damageSources().cactus(), 4);
        handle.hurtServer(handle.serverLevel(), handle.level().damageSources().cactus(), 4);
        return true;
    }

    @EventHandler
    public void on(EntityDamageEvent event) {
        getLogger().info("Invulnerability reduction before: " + event.getDamage(EntityDamageEvent.DamageModifier.INVULNERABILITY_REDUCTION));
        event.setDamage(3.2);
        getLogger().info("Invulnerability reduction after: " + event.getDamage(EntityDamageEvent.DamageModifier.INVULNERABILITY_REDUCTION));
    }
}

Without this fix:

[TestPlugin] Invulnerability reduction before: 0.0
[TestPlugin] Invulnerability reduction after: 0.0
[TestPlugin] Invulnerability reduction before: -3.200000047683716
[TestPlugin] Invulnerability reduction after: 0.0

With this fix:

[TestPlugin] Invulnerability reduction before: 0.0
[TestPlugin] Invulnerability reduction after: 0.0
[TestPlugin] Invulnerability reduction before: -3.200000047683716
[TestPlugin] Invulnerability reduction after: -3.200000047683716

cc @lynxplay

[Lulu13022002]

This computation is flawed anyway since it doesn't take in account the recent damage modifier added by Spigot (freezing) accounted in Vanilla. So the armor generally take less damage in some environment.

For example with a datapack that clear the #bypasses_armor damage type tag and add the player to the #freeze_hurts_extra_types entity type tag.