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/frank-default-small.env

@@ -39,9 +39,11 @@ ALIAS=frank-default-small
 BIND_ADDR=0.0.0.0
 LISTEN_PORT=52625
 
-# Serving defaults from the model card: 64K context, single slot, f16 KV,
-# context checkpoints, prompt cache. Native MTP depth-4.
-CTX_SIZE=65536
+# Serving: 128K context (user request, 2026-08-18; model card default is
+# 64K), single slot, f16 KV, context checkpoints, prompt cache. Native MTP
+# depth-4. KV at 128K is 8 GiB (64 KiB/token); estimated total ~43 GiB of
+# the 64 GiB SVM pool, leaving ~21 GiB headroom.
+CTX_SIZE=131072
 PARALLEL=1
 BATCH_SIZE=2048
 UBATCH_SIZE=2048
@@ -60,6 +62,10 @@ SPEC_DRAFT_P_SPLIT=0.10
 TEMPERATURE=0.7
 TOP_P=0.95
 
-REASONING=off
+# Thinking enabled (user request, 2026-08-18). deepseek format moves the
+# <think> reasoning into message.reasoning_content (OpenAI-compatible);
+# none would leave the thought tags unparsed in content.
+REASONING=on
+REASONING_FORMAT=deepseek
 
 VERBOSITY=4