fix: HOTFIX prevent AUTODETECT model from being queried on Ollama

[fbricon]

bf8b4bd introduced a regression which makes Ollama.ts query /api/show for AUTODETECT models. Since it's not a real model, the log is littered with:

The model "AUTODETECT" was not found. To download it, run ollama run AUTODETECT.

Code: undefined
Error number: undefined
Syscall: undefined
Type: undefined

Error: The model "AUTODETECT" was not found. To download it, run ollama run AUTODETECT.
at _Ollama.parseError (/Users/fbricon/Dev/souk/continue-for-granite/core/llm/index.ts:421:16)
at processTicksAndRejections (node:internal/process/task_queues:105:5)
at customFetch2 (/Users/fbricon/Dev/souk/continue-for-granite/core/llm/index.ts:468:25)
at withExponentialBackoff (/Users/fbricon/Dev/souk/continue-for-granite/core/util/withExponentialBackoff.ts:14:22)

While AUTODETECT model is not a real model, models satisfying isFromAutoDetect are though. While looking for fimSupport is probably not relevant now for autodetected models, I'm considering applying the same sort of logic to detect if thinking is supported, so /api/show would still need to be queried for those models. But that's another story.

cc @RomneyDa

---

Summary by cubic

Prevent Ollama from calling /api/show for the sentinel AUTODETECT model to stop “model not found” log spam. The constructor now early-returns only when options.model === "AUTODETECT", so real auto-detected models continue to be queried normally.

[RomneyDa]

@fbricon nice find, thanks for fixing my mistake

[sestinj]

🎉 This PR is included in version 1.13.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀