fix(validation): tell the CAPTURE it is a PyroWave session, or NVIDIA hands it a CUDA buffer

`--codec pyrowave` selects the ENCODER. The capture pipeline picks its consumer from
`ZeroCopyPolicy::pyrowave_session`, which on the spike path is fed only by the global
`PUNKTFUNK_ENCODER=pyrowave` lab lever (punktfunk-host/src/capture.rs). Without it, .21 resolved

    capture pipeline resolved: cuda-import -> nvenc   capture_arm="cuda-import" consumer="nvenc"
    zero-copy: dmabuf imported to CUDA (no CPU copy)  nv12=true

and the wavelet encoder refused the payload on its first submit: "unsupported FramePayload (need
Dmabuf or Cpu RGB)". That is not a worker bug — the arm that failed was the pure in-process one.

It reproduces only where the A/B actually lives. An AMD box has no CUDA arm to pick, so .25 resolved
straight to dmabuf-passthrough and the kit looked correct there. With the lever set, .21 resolves
`dmabuf-passthrough -> pyrowave` and both arms encode 2700/2700 frames.

V3b then passes on .21 (RTX 5070 Ti, GRID 2 at ~100% GPU, 5120x1440 — the portal captures the real
monitor, --width/--height being synthetic-only):

    in-process, refused       p50 2.85 ms   p99 8.39 ms   (10 windows)
    capped worker, granted    p50 2.65 ms   p99 4.10 ms   (11 windows)
    p99 delta -4.29 ms

The worker reports `priority=Granted(Realtime)` with `ext=VK_KHR_global_priority` on the FIRST
attempt and logs no fallback line; the refused arm logs "every global queue priority class was
refused". So the capability still buys the lever from a SEPARATE process, with the IPC hop in the
loop — 8.39 -> 4.10 ms is a 51% p99 cut, against PW1's in-host 6.4 -> 4.4 at 1080p. Different
resolution and a harder load, so treat the class as confirmed and the absolute numbers as not
comparable to PW1's.

scripts/validate-encode-worker.sh

@@ -527,7 +527,15 @@ run_spike() {
   # NO_COLOR: keep the captured log plain at the source. `log_cat` strips SGR escapes anyway, but a
   # plain log is also the one a human greps by hand when a leg goes red, and `p99_us\e[0m\e[2m=…`
   # defeats the obvious grep just as thoroughly as it defeated this kit's parser.
-  local -a envs=(PUNKTFUNK_PERF=1 NO_COLOR=1)
+  #
+  # PUNKTFUNK_ENCODER=pyrowave is NOT redundant with `--codec pyrowave`, and leaving it out cost a
+  # real V3b run on .21. `--codec` selects the ENCODER; the capture pipeline picks its consumer from
+  # `ZeroCopyPolicy::pyrowave_session`, which on the spike path is fed only by this env (see
+  # punktfunk-host/src/capture.rs — "the global PUNKTFUNK_ENCODER=pyrowave lab lever"). Without it an
+  # NVIDIA host resolves `cuda-import -> nvenc`, imports the dmabuf to CUDA as NV12, and hands the
+  # wavelet encoder a payload it rejects: "unsupported FramePayload (need Dmabuf or Cpu RGB)". AMD
+  # boxes hide this — they have no CUDA arm to pick — so it reproduces only where the A/B lives.
+  local -a envs=(PUNKTFUNK_PERF=1 NO_COLOR=1 PUNKTFUNK_ENCODER=pyrowave)
   for kv in "$@"; do envs+=("$kv"); done
   local -a cmd=("$HOST_BIN" spike --codec pyrowave --source "$SPIKE_SOURCE"
                 --width "$OPT_WIDTH" --height "$OPT_HEIGHT" --fps "$OPT_FPS"