Nebius provider uses retired endpoint and stale model aliases

[demianarc]

Relevant environment info

• Continue version: current main
• Provider: nebius

Description

Continue's existing Nebius provider still points to the retired Nebius AI Studio host (https://api.studio.nebius.ai/v1/) in both the core provider and the OpenAI adapter. The provider UI and docs also use the former “Nebius AI Studio” branding and old console links.

The provider additionally rewrites several friendly model aliases through a static map. That catalog has drifted (and one alias, llama3.1-8b, currently resolves to the 70B model), so valid user-supplied catalog IDs can be changed incorrectly or routed to retired IDs.

Nebius now documents Token Factory at https://api.tokenfactory.nebius.com/v1/, with model discovery through the authenticated /v1/models endpoint.

Expected behavior

• The existing nebius provider uses the documented Token Factory base URL.
• User-configured model IDs are passed through unchanged so they can track the live provider catalog.
• Provider UI/docs use current Token Factory naming and links without unverified pricing/trial claims.
• Focused regression coverage locks down the endpoint and OpenAI-compatible chat/stream/tool request behavior.

I can submit a focused PR with these corrections and tests.