Say what was hit, which way is out, and how far

The narrowphase has been producing a contact since it went in, and the interpreter
was throwing it away. `RCOLLISION(n, f)` reports it: what was hit (a sprite or a
`SOLID` rectangle), which one, the contact normal, the penetration depth, the
contact point, and which axis to reverse.

**The normal points out of the other thing and toward this one**, so a program
moves along it by the depth and is exactly clear. That sign is the one assertion
in the new test that could not be caught any other way -- both parties of a
sprite-against-sprite hit get their own record, each pointing the way *that*
sprite has to move, and sharing one would tell both to go the same direction,
which is how two things end up stuck inside each other.

**Field 7 is the one that deletes the most BASIC.** It is the minimum translation
axis, computed from the normal in C, and it is there because doing it in BASIC
means comparing two floats -- which is exactly where this dialect's left-operand
rule catches people. `BALLBRICKS`/`TESTCELL` in the artwork breakout spend six
lines computing an overlap rectangle and comparing its width to its height to get
this number.

The record is **sticky and deepest-wins**: replaced whenever that sprite is in a
contact and otherwise left alone, so `BUMP` stays the event and this stays the
detail of it. Making it clear itself when nothing touches would break the pairing,
because `BUMP` accumulates across steps and a once-a-frame poll would find the
detail already gone. Reading `BUMP` clears both, so they cannot disagree.

Deliberately narrower than `akgl_Contact`: no actor pointers, because BASIC has no
actor; no tile fields, because there is no tilemap; no z, because every test is
planar; and **no `dt` and no `sensor`**, which libakgl documents as filled in by
the resolver. This interpreter never resolves anything, so those two come back
zero and mean nothing, and an always-zero field in a reference table is a lie.

Documented with the two caveats that matter: fields 2, 3 and 4 are floats and want
a `%` variable, and the contact *point* is exact only for two boxes -- libakgl's
solver returns a point on the portal it converged to, while the normal and depth
are exact for every pair.

Chapter 8's collision section stops claiming only type 1 exists, which has been
false since the previous commit.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwxGB6TdoVvZ11KQQME9cL

TODO.md

@@ -2220,7 +2220,12 @@ should loop — and the third is why both games are built out of `LABEL` and `GO
 `akgl_CollisionShape` and a pooled proxy. Three things were deliberately left for later, and
 each is here so the reasoning does not have to be reconstructed.
 
-38. **Finish moving `COLLISION` and `BUMP` onto the subsystem.** What landed is the narrowphase
+38. **Finish moving `COLLISION` and `BUMP` onto the subsystem.** **Half done.** The contact
+    is no longer discarded: `RCOLLISION(n, f)` reports what was hit, the normal, the depth,
+    the contact point and the minimum translation axis, and
+    `tests/akgl_backends.c::test_contact_geometry` pins the normal's *sign* for both parties
+    of a sprite-against-sprite hit -- getting that backwards tells both to move the same way,
+    and nothing else would have noticed. What stands is the pairing and the threshold below. What landed is the narrowphase
     and the shapes; the seam is still `collisions(self, uint16_t *mask)` and the pairing is
     still all-against-all over eight slots with a bounding-box reject in front of it.