feat: add Kanban task reasoning overrides

[latentoperator]

Summary

Adds per-task reasoning effort overrides for Hermes Kanban workers.

• stores an optional reasoning_override on Kanban tasks
• accepts --reasoning on hermes kanban create and /kanban create
• exposes reasoning_override on the kanban_create tool
• dispatches workers with a top-level --reasoning <level> CLI flag, mirroring -m/--model, instead of adding a non-secret HERMES_* env var
• threads top-level hermes --reasoning ... through chat/TUI invocation paths

Verification

• python -m pytest tests/hermes_cli/test_kanban_reasoning_override.py tests/hermes_cli/test_argparse_flag_propagation.py -o 'addopts=' -q → 24 passed
• parser smoke: top-level --reasoning before chat, chat-level --reasoning, and -z/--oneshot parsing all preserve the requested effort
• CLI help smoke: python -m hermes_cli.main --reasoning high --version and python -m hermes_cli.main chat --reasoning high --help

I also ran python -m pytest tests/hermes_cli/test_kanban*.py -o 'addopts=' -q; it currently reports 659 passed / 10 failed. The failing cases pass when run as the isolated failing subset, so this appears to be pre-existing order/global-state coupling in the broader Kanban suite rather than this patch. The failures are in crash/reap/decompose tests unrelated to the reasoning override path.

[latentoperator]

The reasoning behind this PR: I tend to use frontier models for planning/orchestration and then route implementation and review to other models via the Kanban board. I've run into several instances where I really needed to dial in the reasoning level depending on the task and the model being used. The only workaround I've found is to change the default model on the agent profile temporarily.

[teknium1]

Thanks for the thoughtful Kanban workflow and for avoiding a non-secret environment-variable setting.

This automated hermes-sweeper review is closing this under the standing delegation-model-routing policy: Hermes does not add per-task model-complexity routing controls, including task-specific reasoning-depth selection. The supported configuration direction is profile/config-based reasoning selection (agent.reasoning_effort), rather than persisting an override on each Kanban card.

• The proposed task field and dispatch argument are in hermes_cli/kanban_db.py (PR commit 5779b76b8ba0).
• The same routing control is exposed to agents via kanban_create in tools/kanban_tools.py (PR commit 5779b76b8ba0).
• Current configuration documents supported profile-level reasoning values in website/docs/user-guide/configuration.md:1273.

Thank you again for the implementation and tests.

---

Closed as not-planned per standing maintainer policy (delegation-model-routing). This is a design-direction decision, not a code-quality judgment — see the Contribution Rubric in AGENTS.md for what the project is looking for. If you believe this policy was misapplied to your change, comment here and a maintainer will take a look.