tune(vllm): pin --enable-prefix-caching on granite + qwen

Benched granite prefix caching at ~6.5x faster TTFT (45ms cached vs 292ms
uncached) on a shared ~4.5k-token summarizer template. granite already had it
on by vLLM-v1 default; pinned explicit so a version flip can't silently disable
it. qwen (nightly) defaulted it OFF -> flipped on (free for the text-chat path,
marginal for vision where each image is a distinct prefix). Soft/evictable KV,
zero memory change (GPU1 still ~3.7GB free), all 5 services healthy.

stacks/qwen35-vl/compose.yaml

@@ -54,6 +54,9 @@ services:
       - ${QWEN_MAX_MODEL_LEN}
       - --dtype
       - auto
+      # Prefix caching pinned ON (the nightly defaults it OFF). Free win for the
+      # text-chat path; marginal for vision (each image is a distinct prefix).
+      - --enable-prefix-caching
     deploy:
       resources:
         reservations:

stacks/vllm/compose.yaml

@@ -243,6 +243,11 @@ services:
       # FP8 KV cache — halves KV memory; near-lossless on Ada (cc 8.9).
       - --kv-cache-dtype
       - ${GRANITE_KV_CACHE_DTYPE}
+      # Prefix caching pinned EXPLICIT (vLLM v1 defaults it on, but pin so a
+      # version flip can't silently disable it). Benched 2026-06-13: ~6.5x faster
+      # TTFT (45ms vs 292ms) on a shared ~4.5k-token summarizer template; soft/
+      # evictable KV, neutral when prefixes don't repeat — pure win for granite.
+      - --enable-prefix-caching
     deploy:
       resources:
         reservations: