test: cover DeepSeek max token limits

[Vasanthdev2004]

What changed

• add regression coverage for OpenAI-compatible provider token caps
• specifically cover deepseek-chat context window and max output token limits
• verify oversized CLAUDE_CODE_MAX_OUTPUT_TOKENS values still clamp to DeepSeek's native 8192 limit

Why

Issue #35 reported deepseek-chat failing with Invalid max_tokens on the DeepSeek OpenAI-compatible endpoint.

The runtime cap logic is already present on main, but it was easy for this behavior to look unresolved without focused coverage. This PR locks the behavior down so provider-specific limits for DeepSeek stay protected against future regressions.

Impact

• helps prevent regressions on DeepSeek and other OpenAI-compatible provider cap handling
• gives maintainers a focused test path for the reported failure mode
• makes issue triage clearer because the expected DeepSeek limit is now encoded in tests

Validation

• bun test src/utils/context.test.ts
• bun run test:provider