fix(skills): make weather workflow Windows-safe

[KailBug]

Problem

On Windows PowerShell, bare curl may resolve to the Invoke-WebRequest alias instead of the native cURL executable.

The weather skill previously used bare curl in its examples. This could cause the first weather command to fail, requiring the agent to analyze the error and retry with curl.exe. Requests for today's weather could also trigger separate current-condition and daily-forecast calls.

These retries introduce avoidable tool calls, LLM iterations, latency, and token usage.

Changes

• use curl.exe for weather requests on Windows PowerShell
• document the platform-specific difference between Windows PowerShell and macOS/Linux shells
• use explicit HTTPS wttr.in URLs
• instruct the agent to use a single ?1&m request for today's weather
• add a regression test for the bundled weather skill guidance

Result

Manual verification with “北京今天天气怎么样？” confirmed that the agent directly called:

curl.exe -s "https://wttr.in/Beijing?1&m"

This avoids the failed bare curl attempt, the follow-up recovery iteration, and a redundant separate current-weather request, reducing avoidable response latency and token usage.

Testing

• pytest tests/agent/test_builtin_weather_skill.py -v — 1 passed
• pytest tests/agent/test_context_builder.py -q — 49 passed
• ruff check tests/agent/test_builtin_weather_skill.py — all checks passed
• manually verified the workflow on Windows PowerShell with:
  nanobot agent --logs -s cli:weather-skill-test
• confirmed that the agent directly called:
  curl.exe -s "https://wttr.in/Beijing?1&m"
• confirmed that no failed bare curl attempt or redundant current-weather request occurred

[Re-bin]

LGTM. The built-in weather workflow now uses Windows-safe PowerShell commands, HTTPS assets, and platform-neutral output guidance, backed by focused skill regression tests.

Verified against the current main; all required checks pass.