[FEATURE]: Bump gitlab-ai-provider to 6.15.1 on v2 (Fable 5.1, GPT-6 Astra, workflow reconnect fix) #50319

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

The v2 branch still pins gitlab-ai-provider at 6.12.1 in packages/core/package.json, while the latest published version is 6.15.1. The bumps to 6.13.0 (#46913) and 6.15.0 (#47791) landed on dev but never made it onto v2, so v2 is missing five releases' worth of model mappings and fixes.

Bumping v2 from 6.12.1 to 6.15.1 picks up:

• 6.13.0 — Claude Fable 5.1 model mapping (duo-chat-fable-5-1).
• 6.15.0 — GPT-6 Astra model mapping (duo-chat-gpt-6-astra, routed via the Responses API).
• 6.14.0 — configurable reasoning controls: a thinking option for the Anthropic Messages models (including the adaptive mode that effort-based models such as Claude Opus 4.8 require) and a reasoningEffort option for the OpenAI models.
• 6.12.2 — waits out the workflow lock before reconnecting after an approval, fixing a race with Workhorse that rejected reconnects with close code 1013.
• 6.15.1 — scopes Agent Platform model discovery by projectId (which works for both group- and user-namespace projects) with rootNamespaceId retained as a fallback for older GitLab instances, and corrects stale workflow refs.

Benefit: GitLab Duo users on v2 get the two current flagship models in the model picker, the agentic workflow reconnect bug goes away, and model discovery starts working for projects in personal namespaces. The API surface change is purely additive — discover, getEffectiveModelRef, and discoverModels still accept a bare namespace string — so this is a version bump only with no code changes needed in packages/core/src/plugin/provider/gitlab.ts.