feat(fetch): make the request timeout configurable

[ameyypawar]

Summary

The fetch server hardcoded a 30s httpx timeout with no way to override it — too short for large downloads and slow endpoints, too long for quick health checks (#4448).

The content-fetch timeout is now configurable, most-specific wins:

• per-request: an optional timeout argument on the fetch tool
• server default: the --timeout CLI flag or the FETCH_TIMEOUT environment variable
• fallback: the existing 30s default

Change

• server.py: thread a timeout parameter through fetch_url and serve, and add an optional timeout field (gt=0) to the Fetch tool-argument model. call_tool resolves the effective timeout (per-request overrides the server default).
• __init__.py: add the --timeout flag and FETCH_TIMEOUT env var; both reject non-positive values (matching the tool field's gt=0) with a clean error.
• Tests: fetch_url applies the default/explicit timeout; --timeout, FETCH_TIMEOUT, and their precedence.
• README: document the flag, the env var, and the per-request argument.

Notes

• Unit is seconds (float), not the milliseconds suggested in the issue — this matches httpx's timeout unit and the previous hardcoded timeout=30, and is idiomatic for a Python/httpx server. Happy to switch to milliseconds if maintainers prefer the issue's original convention.
• Scope: only the content fetch is affected. The robots.txt preflight keeps its existing default timeout, so its behavior is unchanged.

Verified locally: uv run pytest → 26 passed, uv run pyright → 0 errors, uv run ruff check → clean, uv build → ok.

Addresses #4448.