fix(opencode): fix 1M context for Opus - compaction token counting (#154)

[randomm]

Summary

Fixes Opus hitting 200k token limit despite context-1m header.

Changes

• Compaction fix: Remove output tokens from count - API only enforces limit on input tokens
• Context override: Anthropic Sonnet/Opus models get 1M context limit
• Preserved: adaptive-thinking header and effort-based thinking for Opus 4.6

Also includes

• New /upstream-gap custom command for analyzing fork divergence

Fixes #154

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.