[FEATURE] Expose 1M-token context window in Cowork (currently 200K cap)

Problem Statement

Cowork caps context at 200K tokens — the same limit as the web/chat interface. The 1M-token context window is available on the API (Claude Opus 4.8 / Sonnet 4.6 / Haiku 4.5), Bedrock, and Vertex, but is not exposed in the consumer surfaces of Chat and Cowork.

For agentic, multi-file, long-session work — exactly the use case Cowork is designed for — 200K is frequently a bottleneck. Tasks like large file analysis, long document processing, or extended multi-step sessions hit the limit in ways that Chat and API usage do not.

Desired Behavior

Cowork should allow users to select a 1M-token context window (where the underlying model supports it), either as a per-session setting in the model/effort selector, or as a global Cowork preference in Settings.

Why This Matters for Cowork Specifically

Cowork is the agentic surface — it reads files, runs code, maintains long plans. It needs larger context more than any other surface. Keeping it at 200K while the API exposes 1M is the wrong trade-off for the product primary use case.

Alternatives Considered

• Using the API directly — loses the Cowork UI, file access, tool integrations
• Splitting tasks — workaround, not a fix; breaks coherence for long sessions