fix: honor opencode-go max reasoning effort

[danshapiro]

Summary

• accept max as a reasoning effort level in config and gateway command handling
• have the opencode-go provider profile send deepseek-v4-* reasoning effort as a top-level Chat Completions reasoning_effort parameter
• keep non-DeepSeek OpenCode Go models unchanged and avoid generic extra_body.reasoning for the DeepSeek V4 profile path

Tests

• uv run --frozen --extra dev pytest tests/providers/test_provider_profiles.py::TestOpenCodeGoProfile tests/providers/test_transport_parity.py::TestOpenCodeGoParity tests/gateway/test_reasoning_command.py::TestReasoningCommand::test_parse_reasoning_command_args_accepts_ascii_and_smart_global_flags
• uv run --frozen --extra dev pytest tests/providers/test_provider_profiles.py tests/providers/test_transport_parity.py tests/gateway/test_reasoning_command.py
• git diff --check

[teknium1]

Thanks for the focused OpenCode Go regression coverage. The provider/config portion of this change is already present on current main, but the gateway command portion remains useful and needs a manual salvage.

Problems

• Current /reasoning execution lives in gateway/slash_commands.py:2634, not the old gateway/run.py location changed by this PR. Its accepted-level set at gateway/slash_commands.py:2732 still omits max, so the gateway defect remains while the branch is conflicting.
• Current main's OpenCodeGoProfile already maps max to top-level DeepSeek reasoning_effort at plugins/model-providers/opencode-zen/__init__.py:115-125, including prefixed model IDs covered by tests/plugins/model_providers/test_opencode_go_profile.py:115-123. Replacing it with the older profile would discard newer Kimi/GLM and normalization behavior.

Suggested changes

• Transplant the gateway acceptance change into gateway/slash_commands.py:2732 and add max to hermes_cli/commands.py:155-157.
• Add direct session and --global /reasoning max tests; tests/gateway/test_reasoning_command.py:84-87 currently checks only argument parsing.

This is an automated hermes-sweeper review.

[teknium1]

Closing — fully superseded on current main. max is accepted in config/gateway handling (#62650, strings completed in #65651), and the opencode-zen profile now sends DeepSeek reasoning effort as a top-level reasoning_effort Chat Completions param with max/ultra clamping (see plugins/model-providers/opencode-zen/__init__.py). Thanks for the contribution!