docs: add QVAC local provider

[simon-iribarren]

Issue for this PR

No linked issue — documentation addition.

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Adds a ### QVAC section to the providers docs, next to the existing llama.cpp / LM Studio / Ollama local-provider entries.

QVAC is a local-first AI runtime that exposes an OpenAI-compatible server via qvac serve openai; the @qvac/ai-sdk-provider npm package is a branded wrapper around @ai-sdk/openai-compatible that points at it. The section shows the opencode.json provider block plus the matching qvac.config.json, and calls out the non-obvious things needed for it to behave as an agent backend:

• preload the model alias referenced from opencode.json;
• set ctx_size explicitly because the default 1024-token context is too small for coding-agent prompts;
• use a capable local model for tool-calling quality, with gpt-oss-20b as the recommended local backend.

QVAC now queues same-model completion requests, so model and small_model can point at the same local alias. A separate lighter small_model remains optional if users want title, summary, or compaction calls to avoid waiting behind long chat responses.

A QVAC provider entry is also in review at anomalyco/models.dev#1954, so QVAC will show up in /connect once that lands; this docs section is the explicit equivalent.

How did you verify your code works?

Ran qvac serve openai locally and pointed OpenCode at it using this provider config. The single-model setup now works for both the main chat model and OpenCode's small_model utility calls because QVAC queues same-model completion requests.

Screenshots / recordings

Docs-only change.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR