Should fix https://github.com/open-goal/jak-project/issues/2679
Here's a test program that will trigger the bug when near these guards:
```lisp
(define *cquery* (new 'global 'collide-query))
(defun test-bad-collide ()
(let ((lower (new 'static 'vector :x 1681893.8750 :y 61314.2031 :z 345208.6562 :w 1.))
(upper (new 'static 'vector :x 1701603.8750 :y 67624.0625 :z 357881.0312 :w 1.))
;(cquery (new 'stack-no-clear 'collide-query))
)
(set! (-> *cquery* collide-with) (the-as collide-spec 1))
(set! (-> *cquery* ignore-process0) #f)
(set! (-> *cquery* ignore-process1) #f)
(set! (-> *cquery* ignore-pat) (new 'static 'pat-surface :noentity #x1 :nojak #x1 :probe #x1 :noendlessfall #x1))
(set! (-> *cquery* action-mask) (collide-action solid))
(set! (-> *cquery* bbox min quad) (-> lower quad))
(set! (-> *cquery* bbox max quad) (-> upper quad))
(format 0 "doing collide...~%")
(fill-using-bounding-box *collide-cache* *cquery*)
(format 0 "have ~d and ~d~%" (-> *collide-cache* num-tris) (-> *collide-cache* num-prims))
)
(none)
)
```
As far as I can tell, there's a totally invalid collide-hash with an
inside `axis-scale.z`. On the PS2, it gets ignore because of how
float->int works for floats that are too big. On PC, it ends up using a
negative value and loop forever.
The only interesting one is `collide-hash`, which is untested.
The other two are very likely unused. I skipped the annoying code in
`collide-probe` because it's not used and the same as jak 1.
water111