perf: reduce OLLAMA_NUM_CTX default from 65536 to 16384

65536 was causing Ollama to allocate a ~50GB KV cache, spilling 77% of
the model to CPU RAM and making prefill extremely slow (35-125s TTFT).

16384 covers 30+ message conversations comfortably while keeping the KV
cache small enough to stay on GPU.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/fabledassistant/config.py

@@ -24,9 +24,10 @@ class Config:
     )
     OLLAMA_URL: str = os.environ.get("OLLAMA_URL", "http://localhost:11434")
     OLLAMA_MODEL: str = os.environ.get("OLLAMA_MODEL", "qwen3:latest")
-    # KV cache context window for generation. Higher = more RAM usage but longer inputs/outputs.
-    # 131072 is the practical maximum for most models. Lower this on RAM-constrained hosts.
-    OLLAMA_NUM_CTX: int = int(os.environ.get("OLLAMA_NUM_CTX", "65536"))
+    # KV cache context window for generation. Keep this as small as practical —
+    # a larger context forces more KV cache into CPU RAM, drastically slowing prefill.
+    # 16384 covers ~30+ message conversations with our system prompt comfortably.
+    OLLAMA_NUM_CTX: int = int(os.environ.get("OLLAMA_NUM_CTX", "16384"))
     SECRET_KEY: str = _read_secret("SECRET_KEY", "SECRET_KEY_FILE", "dev-secret-change-me")
     SECURE_COOKIES: bool = os.environ.get("SECURE_COOKIES", "").lower() in ("1", "true", "yes")
     LOG_LEVEL: str = os.environ.get("LOG_LEVEL", "INFO")