feat(provider): graceful context-1m fallback for sub-Tier-4 accounts

[sjawhar]

Summary

• Adds graceful error handling when the context-1m-2025-08-07 Anthropic beta header is rejected by accounts below Tier 4
• On the first request, if Anthropic returns HTTP 400 with "long context beta not available", the fetch wrapper strips the header, retries immediately, and remembers to skip it for the rest of the process
• Zero configuration required — Tier 4+ accounts get 1M context automatically, sub-Tier-4 accounts fall back to 200K seamlessly

Changes

• packages/opencode/src/provider/provider.ts — Added _context1mDisabled flag and stripContext1m() helper in the fetch wrapper for error-based retry
• packages/opencode/test/provider/context1m.test.ts — 9 tests covering header stripping and error detection

Context

The feat/opus-4-6-adaptive-thinking branch adds context-1m-2025-08-07 to the static Anthropic beta header, enabling 1M context for supported models. However, this breaks accounts below Tier 4 ($400 cumulative deposits) which receive a hard error. This PR makes the feature degrade gracefully.

Note: The context-1m header alone does NOT change pricing — premium rates (2× input, 1.5× output) only apply when actual input tokens exceed 200K.

[github-actions]

This PR doesn't fully meet our contributing guidelines and PR template.

What needs to be fixed:

• PR description is missing required template sections. Please use the PR template.

Please edit this PR description to address the above within 2 hours, or it will be automatically closed.

If you believe this was flagged incorrectly, please let a maintainer know.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

I found some related PRs, but none appear to be duplicates of PR #15641. Here's what I found:

Related (but distinct) PRs:

1. feat(provider): add 1M context window support for Anthropic models (PR feat(provider): add 1M context window support for Anthropic models #14375)

   • Likely implements the initial 1M context feature that PR feat(provider): graceful context-1m fallback for sub-Tier-4 accounts #15641 builds upon

2. feat(provider): add adaptive thinking and 1M context support for Claude Opus 4.6 (PR feat(provider): add adaptive thinking and 1M context support for Claude Opus 4.6 #12342)

   • Earlier work on context support, likely a foundation for this feature

3. fix(provider): enable 1M context for Vertex AI Anthropic models (PR fix(provider): enable 1M context for Vertex AI Anthropic models #14055)

   • Related to context features but for Vertex AI specifically

PR #15641 is distinct from these because it specifically addresses the graceful degradation for sub-Tier-4 accounts (adding error handling and fallback logic) rather than implementing the core 1M context feature itself.

No duplicate PRs found

[sjawhar]

Closing — this PR includes unrelated merge parents. The context-1m fallback change will be included when the parent feature branch (feat/opus-4-6-adaptive-thinking) is ready.