#1405 tools: add configurable launches and reuse local model downloads

Descrição

## Summary

After installation, keep everyday launch settings in `strata.yaml` and start with `make run`. Model selection, context size, server settings and API keys can be configured while inheriting the engine paths and hardware choices saved by setup.

Setup reuses complete local GGUF files, including files without a `.done` marker. Replacing model files requires `--force-download`.

## What changed

- Add `Makefile`, `run.py`, `serve/launchconfig.py` and `strata.example.yaml`, with `init`, `models`, `check` and `run` commands.
- Support YAML configs in the server and web Settings view. Default to loopback, require an API key for network access, and support keys supplied through environment variables. Settings edits update the YAML overlay.
- Preserve incomplete existing GGUF files and resume missing files from `.part`. Forced downloads transfer through `.part`, replace completed files and rebuild the prepared model pack.
- Document the launch and replacement workflows, exclude local config files from Git and Docker builds, and add regression coverage.

## Extra Notes

- 228 tests passed across launch, setup, download, shard and engine checks. Coverage includes a mock HTTP server, authentication, Settings persistence, download reuse and forced replacement.
- Python compilation and `git diff --check` passed.
- GPU inference was not tested. `make check` validates settings and required files; it does not determine whether the chosen context fits available memory.