fix(llm): resolve context window via max_input_tokens with 128k fallback for unmapped models

- get_context_window() now reads litellm.get_model_info().max_input_tokens
  instead of get_max_tokens(), which returns the output cap and severely
  under-reported the window for every mapped model (e.g. deepseek 8k vs 1M)
- models litellm cannot map fall back to 128000 with a single warning per
  model instead of warning on every call and returning None
- raise LLMConfig.max_tokens default 8192 -> 32768 to match the template
- README: configure the API key directly in llm_config.yaml; document
  max_tokens / context_window in the example

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

opc/core/config.py

@@ -272,12 +272,13 @@ class LLMConfig(BaseModel):
     routing: dict[str, str] = Field(default_factory=dict)
     fallback: dict[str, Any] = Field(default_factory=dict)
     temperature: float = 0.3
-    max_tokens: int = 8192
+    max_tokens: int = 32768
     # Total input context window (tokens) for the active model. Set this when
     # the model is not mapped in litellm (e.g. proxy/self-hosted models like
     # doubao/minimax/glm), so the context-usage ring and compaction thresholds
-    # have a real denominator. 0 = auto-detect via litellm. Optional per-model
-    # overrides keyed by model name take precedence over the scalar value.
+    # have a real denominator. 0 = auto-detect via litellm; unmapped models
+    # fall back to 128000. Optional per-model overrides keyed by model name
+    # take precedence over the scalar value.
     context_window: int = 0
     context_window_overrides: dict[str, int] = Field(default_factory=dict)