[Feature]: Bundled llama.cpp cannot load Spark-X2.5 GGUF — unknown model architecture 'spark2_5'

## QwenPaw Version

2.2.0b5 (`qwenpaw --version`, confirmed 2026-09-01), QwenPaw Desktop.

## Description

Downloaded `XHToken/Spark-X2.5-4B-GGUF` (iFlytek Spark-X2.5, GGUF V3, BF16, 7.66 GiB) from ModelScope via the QwenPaw Desktop local-model library. QwenPaw then tries to launch the bundled llama-server twice; both attempts exit during model load, and the UI shows:

> llama.cpp server exited before becoming ready - {"detail":"llama.cpp server exited before becoming ready"}

The bundled llama.cpp build does not recognize the new `spark2_5` model architecture.

**Related PR(s):** llama.cpp upstream support for Spark-X2.5 is still unmerged: ggml-org/llama.cpp#27868 (draft, opened 2026-08-28)

**Security considerations:** None.

## Component(s) Affected

- [x] Core / Backend (app, agents, config, providers, utils, local_models)

## Environment

- **QwenPaw version:** 2.2.0b5
- **OS:** Windows 10 (AMD64)
- **Install method:** QwenPaw Desktop
- **Python version (if applicable):** bundled runtime

## Steps to Reproduce

1. Open the local-model library in QwenPaw Desktop (2.2.0b5, Windows 10).
2. Download `XHToken/Spark-X2.5-4B-GGUF` (7.66 GiB, GGUF V3, BF16) — download completes normally.
3. Launch the model → llama-server exits before becoming ready (UI error toast; two launch attempts, both failed).

## Actual vs Expected

- **Actual:** llama-server exits during model load with `unknown model architecture: 'spark2_5'`; UI shows "llama.cpp server exited before becoming ready".
- **Expected:** Once llama.cpp upstream support is merged and released, the bundled llama.cpp should be upgraded so `spark2_5` GGUF models (e.g. Spark-X2.5-4B) can run from the local-model library.

## Logs / Screenshots

`llama-server.log` (2026-09-01, model: Spark-X2.5-4B.gguf):

```
llama_model_loader: - type f32: 73 tensors
llama_model_loader: - type bf16: 217 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = BF16
print_info: file size = 7.66 GiB (16.00 BPW)
llama_model_load: error loading model: error loading model architecture: unknown model architecture: 'spark2_5'
llama_model_load_from_file_impl: failed to load model
main: exiting due to model loading error
```

## Additional Notes

- Upstream llama.cpp support for Spark-X2.5 is still a draft PR: ggml-org/llama.cpp#27868. Once it is merged and released, please consider upgrading the bundled llama.cpp (llama-server.exe) so Spark-X2.5 GGUF works in the local-model library.
- Similar past case: #823 (Qwen3.5 architecture; resolved by upgrading the bundled llama.cpp).
- Workaround: switch to a supported local model (e.g. `QwenPaw-Flash-4B-Q4_K_M` starts fine).

---

**Disclosure (per #4333):** This issue was drafted with the help of an AI assistant (a QwenPaw agent). The reproduction steps, version info and logs are real data collected on the reporter's machine (llama-server.log + UI screenshot); the reporter has reviewed the content and vouches for its accuracy.

Start by checking the upstream llama.cpp Spark-X2.5 support PR #27868 and the existing bundled llama-server upgrade path, using issue #823 as a precedent. Once upstream support is merged and released, upgrade the bundled llama-server.exe and verify that XHToken/Spark-X2.5-4B-GGUF loads from the local-model library without the unknown-architecture error.

Vom Indexierungsmodell aus dem Issue-Text verfasst.