Fix: bare aider Connection-refused on WSL + Ollama

On WSL2 with Ollama running on the Windows side, a plain aider invocation (model already set in .aider.conf.yml) failed with OllamaError: … Connection refused even though the model server was healthy. scripts/write_aider_env.py fixes it by writing the endpoint where Aider actually looks.

Why it broke

Aider (via litellm) reads OLLAMA_API_BASE from a repo-root .env, loaded before its own model-info probe. On WSL2, the Windows-side Ollama doesn't answer on the guest's own localhost — only the default-gateway IP (or a resolv.conf nameserver) reaches it. Nothing wrote that reachable address into .env, so Aider kept probing a dead loopback address and retried with backoff until it gave up.

The fix

scripts/write_aider_env.py reuses the reachable-base_url discovery fit_device.py --probe already does (loopback → default-gateway → resolv nameserver — the same fix that shipped for /local-models reports) and writes it into a gitignored project .env:

python scripts/write_aider_env.py --root .            # write/refresh
python scripts/write_aider_env.py --root . --dry-run   # preview only

It's idempotent — re-running after wsl --shutdown or a gateway-IP change updates only the OLLAMA_API_BASE line, leaving the rest of .env (and any unrelated secrets already in there) untouched — and it ensures .gitignore covers .env/.env.* so the machine-local value never gets committed. /local-models's report now points at this tool directly on Aider projects, instead of the previous vague "wire Aider/.env" instruction.

Verified live against Aider 0.86.2 on a real WSL2 host: before the fix, bare aider gave Connection refused / OLLAMA_API_BASE: Not set; after writing .env, the same command resolved model info and began generating with no errors.