docs #429 + probe: frame stalls are pack-independent (~1.7/s both arms); the pack converts them into visible player jumps (59 vs 3); prediction-snap theory dead

Adds the TEMPORARY PlayerPresentationProbe (ACDREAM_PROBE_PLAYER_PRESENT)
per the #429 apparatus plan and records the measured two-arm evidence in
the issue. Probe strips with the fix.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

docs/ISSUES.md

@@ -115,6 +115,39 @@ ACE's >=800 misread), accumulating drift that the AutoPos/correction loop
 snaps back on its cadence. Walking (speed 1.0 both sides) and strafing
 accumulate little or no drift — matching the report exactly.
 
+**MEASURED 2026-08-23 (same session) — the story inverted twice and is now
+pinned by data.** A per-frame probe (`PlayerPresentationProbe`,
+`ACDREAM_PROBE_PLAYER_PRESENT=<csv>`, TEMPORARY, wired in
+`WorldRenderFrameBuilder`/`RuntimeWorldFrameEnvironmentPreparation`)
+captured `(t, camera, presented player position)` per frame over ~50 s
+of running per arm (`artifacts/owner-gate/player-present-429-packON.csv`
+/ `-packoff.csv`, ~270 FPS baseline, median frame 3.7 ms):
+
+- Frame stalls of 15-25 ms (max ~230-260 ms) occur ~1.7/s while moving in
+  BOTH arms — pack ON 88, pack OFF 83 — the Atmospheric pack does NOT
+  cause the stalls. The uncapped run reproducing the hitch had already
+  killed the #235 capped-alias theory.
+- The pack changes what a stall LOOKS like: with pack ON, 59 of the 88
+  long frames carry a 3x-5x player-position jump (the visible hitch);
+  with pack OFF only 3 of 83 do — the same stall lands at a pipeline
+  phase where the presented position has not advanced, so the player
+  stays visually smooth and the owner never perceived it.
+- Right after ON-arm stall clusters, normal-length frames show ~zero
+  player motion (catch-up artifacts). The prediction/AutoPos-correction
+  hypothesis is DEAD: position advances exactly proportionally to
+  elapsed time through the stalls; no snap-back is present in the data.
+
+**Two separated defects:**
+1. BASE CLIENT: periodic 15-25 ms frame stalls (~1.7/s while moving).
+   GC exonerated (Gen0 at 6.3 s cadence, gen1/2 zero). Suspects:
+   streaming publication/upload bursts on movement, present-path stalls.
+   Instrument with a per-stall stack/phase capture, not more theories.
+2. PACK FRAME GRAPH: its ordering shifts the stall's position relative
+   to the physics commit/presented-position sampling, converting silent
+   stalls into visible player jumps. Establish where the ON-arm long
+   frames spend their extra time (the pack's CPU stage profiler names
+   stages) and where the player position is sampled relative to it.
+
 **Next probes (in order):**
 1. `ACDREAM_DUMP_MOTION=1` + a temporary inbound-position log for the
    LOCAL guid: does ACE send position sets for the local player every