[codex] Add K-LLM use-case recipes

[svngoku]

Requirements

Extend the multi-model orchestration path from one hardcoded flow into K-LLM plus tools use cases. The implementation should work locally, include tests, provide concise examples/docs, and remain compatible with the existing multi default behavior.

Changes

• Adds typed use-case recipes for K-worker orchestration in agentic_internet/agents/use_cases.py.
• Adds a tool-bundle runtime resolver in agentic_internet/agents/orchestration_runtime.py.
• Re-expresses the existing research flow as the default research recipe.
• Adds non-search-only proof recipes: technical_due_diligence and market_intelligence.
• Wires agentic-internet multi --use-case ... plus --worker-model worker=model overrides.
• Adds agentic-internet tools --use-cases to list available recipes.
• Updates README examples and architecture docs with the K-LLM recipe lifecycle.
• Adds an exec-plan documenting scope, acceptance criteria, task graph, and verification.

Examples

agentic-internet tools --use-cases
agentic-internet multi "Assess this repo idea" --use-case technical_due_diligence
agentic-internet multi "Analyze this product category" --use-case market_intelligence --worker-model commerce_analyst=gemini-pro

Verification

Ran the full project gate before commit and push:

PATH="$HOME/.local/bin:/opt/homebrew/bin:$PATH" make check

Results:

• Ruff check: passed
• Ruff format check: passed, 50 files already formatted
• Mypy: passed on 31 source files
• Pytest: 196 passed, 4 existing Pydantic deprecation warnings
• Golden principles: passed

Notes

The local git hooks still skipped bead sync because bd is not installed on PATH. No live provider credentials are required for the new unit and CLI tests.