gpu-worker: dispatcher-procs in (health) RPC response

Mirror of pool-procs added earlier — bend health now reports both
pool-procs and dispatcher-procs counts so a single RPC tells the
feeder/factory consumer whether either supervisor is dead while
queue has work.

Failure mode this fixes: 4090's bend-dispatcher hit MAX_IDLE_LOOPS
drain-exit; balance moved 5 cells into its queue but no dispatcher
to consume them. Bend health response previously didn't surface the
gap; feeder's separate SSH probe (now added in www.foxhop.net commit)
caught it but a single RPC is cheaper than per-host SSH.

examples/cuda-fanout/gpu-worker.lsp

@@ -397,6 +397,22 @@
            ((or (eq? line #f) (eof-object? line)) 0)
            (else (or (string->number (parse-trim line)) 0))))))))
 
+(define (health-dispatcher-procs)
+  ;; Count of bend-dispatcher processes on this host. <=1 = dispatcher
+  ;; functionally dead (only stale wrapper bash matched). Live
+  ;; multi-dispatcher = parent + 2 worker forks = ≥3 procs.
+  (let* ((pair (spawn-process-stdio "/usr/bin/pgrep"
+                 '("-cf" "bend-dispatcher"))))
+    (cond
+      ((eq? pair #f) 0)
+      (else
+       (let ((line (read-line (cdr pair))))
+         (close-port (car pair))
+         (close-port (cdr pair))
+         (cond
+           ((or (eq? line #f) (eof-object? line)) 0)
+           (else (or (string->number (parse-trim line)) 0))))))))
+
 (define (parse-trim s)
   ;; Strip leading/trailing whitespace from a single-line string.
   (let* ((n (string-length s))
@@ -439,6 +455,7 @@
         (list 'vram-free-mb (health-vram-free-mb))
         (list 'uptime-ms (current-time-ms))
         (list 'pool-procs (health-pool-procs))
+        (list 'dispatcher-procs (health-dispatcher-procs))
         (list 'queue-ready (health-queue-count "ready"))
         (list 'queue-emitting (health-queue-count "emitting"))
         (list 'queue-done (health-queue-count "done"))))