frank-default-small: 128K context and thinking on

- CTX_SIZE 65536 -> 131072 (user request). KV at 128K is 8 GiB
  (64 KiB/token f16); estimated ~43 GiB used of the 64 GiB SVM pool,
  ~21 GiB headroom.
- REASONING off -> on with REASONING_FORMAT=deepseek, so <think>
  reasoning lands in message.reasoning_content (OpenAI-compatible)
  instead of being left unparsed in content.

scripts/serve-frank-small.sh

@@ -37,7 +37,7 @@ fi
 : "${ALIAS:=frank-default-small}"
 : "${BIND_ADDR:=0.0.0.0}"
 : "${LISTEN_PORT:=52625}"
-: "${CTX_SIZE:=65536}"
+: "${CTX_SIZE:=131072}"
 : "${PARALLEL:=1}"
 : "${BATCH_SIZE:=2048}"
 : "${UBATCH_SIZE:=2048}"
@@ -51,7 +51,8 @@ fi
 : "${SPEC_DRAFT_P_SPLIT:=0.10}"
 : "${TEMPERATURE:=0.7}"
 : "${TOP_P:=0.95}"
-: "${REASONING:=off}"
+: "${REASONING:=on}"
+: "${REASONING_FORMAT:=deepseek}"
 : "${VERBOSITY:=4}"
 
 MODEL="$MODEL_DIR/$MODEL_FILE"
@@ -113,6 +114,6 @@ exec "$LLAMA_BIN" \
     --temp "$TEMPERATURE" \
     --top-p "$TOP_P" \
     --reasoning "$REASONING" \
-    --reasoning-format none \
+    --reasoning-format "$REASONING_FORMAT" \
     --reasoning-budget -1 \
     -lv "$VERBOSITY"