fix(opencode): let recall use server relevance

[arikmeyer]

Summary

• make small coding quota limits exact instead of forcing one slot into every category
• add an optional recallPurpose setting; explicit null omits both the purpose and preset quotas so server relevance can select context
• forward an explicitly configured OpenCode recall token budget to the Context API as max_tokens
• preserve the existing coding default and synchronize the shared recall implementation across agent integrations

This fixes a concrete case where autoRecall.limit: 1 produced six quota slots and injected unrelated categories. OpenCode users can now choose server relevance without disabling automatic recall or query expansion.

Type of Change

• New feature (feat)
• Bug fix (fix)
• Documentation (docs)
• Refactoring (refactor)
• Other

Testing

• node --test examples/memory-plugin-shared/*.test.mjs (62 passed)
• npm test --prefix examples/opencode-plugin (28 passed)
• npm run check --prefix examples/opencode-plugin
• node examples/memory-plugin-shared/sync.mjs
• git diff --check

Checklist

• Code follows project style guidelines
• Tests added for new functionality
• Documentation updated
• All focused tests pass