#11276: feat: change Venice default model to Kimi K2.5

## Summary Change the Venice default model from `llama-3.3-70b` to `kimi-k2-5` (Kimi K2.5). ## Changes - **Default model**: Updated `VENICE_DEFAULT_MODEL_ID` from `llama-3.3-70b` to `kimi-k2-5` - **Model catalog**: Added `kimi-k2-5` entry to the static fallback catalog ## Why Kimi K2.5? Kimi K2.5 is a significantly more capable default model: - **256K context window** (vs 128K for Llama 3.3 70B) - **Vision support** (multimodal text + image input) - **Reasoning support** (built-in chain-of-thought) - **Function calling** compatible - **Code-optimized** See: https://venice.ai/blog/kimi-k2-on-venice ## Resubmission This is a resubmission of #5081 which was closed during feature freeze. ## Testing Minimal changes to static config - CI will verify. <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR updates the Venice integration’s default model from `llama-3.3-70b` to `kimi-k2-5` and adds a static-catalog entry for `kimi-k2-5` so the model list can still be built when Venice’s `/models` discovery endpoint is unavailable. This fits into the existing `discoverVeniceModels()` flow by ensuring `VENICE_MODEL_CATALOG` contains the new default model ID used by `VENICE_DEFAULT_MODEL_REF`, and by allowing the fallback catalog to surface Kimi K2.5 consistently in test environments or when API discovery fails. <h3>Confidence Score: 3/5</h3> - This PR is likely safe to merge, but the default-model change appears to alter the expected privacy mode of the default Venice model. - The change is small and localized to static config, but it switches the default model to one marked `privacy: "anonymized"` in the fallback catalog, which is a real behavioral change for users relying on Venice defaults to be fully private inference. - src/agents/venice-models.ts <!-- greptile_other_comments_section --> <!-- /greptile_comment -->