fix(eff_ob): the packet names an entity and switches an effect on or off - #508

The model has ExtraSpace, MapX, MapY, Unknown, EffectId. The client
reads something else.

Its handler takes the first character of field 0 as the entity kind, field 1
as that entity's id, field 2 as a switch, and field 3 as the effect:

A   := StrToIntDef(field[1], -1)        the entity id
c   := field[0][1]                      the kind - first character only
obj := <lookup>(c, A)
B   := StrToIntDef(field[2], 0)         the switch
E   := StrToIntDef(field[3], 0xFFFF)    the effect
if B = 1:  attach E to obj
if B = 0:  detach whatever obj is carrying
otherwise: nothing

No map coordinate is read anywhere in it.

What was tested

In game, with a positive control so the result cannot be a coincidence - effect
4430 is a ring of magenta arcs and it is unmistakable:

dotnet build: 0 warnings. dotnet test: 127/127. The serializer corpus
snapshot was edited by hand, two lines, so the diff stays two lines.

What this contradicts, and why I still think it is right

SerializeEffObPacketMatchesTrace pinned eff_ob -1 -1 0 4269, and that line
really is in a packet capture - four times. I replaced the assertion rather than
keep it, so this is worth stating plainly.

That capture is of an older protocol. On the modern client the line does
nothing: the double space puts an empty field in front, so everything shifts by
one and field 2 reads -1, which is neither 1 nor 0, so no branch runs. Even
without the shift, -1 names no entity and the lookup fails first. It cannot be
made to draw anything, which is what the fourth row of the table above shows.

If the intent is to keep the model faithful to that capture rather than to the
client that reads it, say so and I will close this.