Deps cleanup PR 2: Replace requests in vespa/models.py

[thomasht86]

Part of #1285.

Background

We want to minimize pyvespa's runtime dependency surface (see #1285). vespa/models.py is one of the last places in vespa/ still importing requests directly.

Scope

Replace requests usage in vespa/models.py:_validate_single_url so this file no longer needs requests.

Current usage

• vespa/models.py:8 — import requests
• vespa/models.py:276 — requests.head(url, timeout=10, allow_redirects=True)
• vespa/models.py:279 — requests.get(url, timeout=10, stream=True, allow_redirects=True) (fallback when HEAD doesn't return 200)
• vespa/models.py:287 — except requests.RequestException

Used to validate user-supplied model_url / tokenizer_url values.

Approach

Recommended: use stdlib urllib.request (one-shot validation, no pooling needed — one fewer thing depending on httpr surface area).

Alternative: use httpr.head / httpr.get (sync API) — fine if httpr is already a runtime dep.

Update the exception handler from requests.RequestException to the new equivalent.

Tasks

• Replace requests.head / requests.get calls
• Update exception handling
• Remove import requests from vespa/models.py
• Unit test covering: HEAD success, HEAD-fails-then-GET-succeeds, both fail → ValueError
  • Recommend pytest-httpserver or a local socket — don't mock the HTTP library directly
• Confirm retry behavior still works (function is wrapped in @retry(stop=stop_after_attempt(3), ...))

Out of scope

• Removing requests from pyproject.toml — that's PR 4 (Deps cleanup PR 4: Remove requests/requests_toolbelt/httpx from pyproject.toml #1289)

Risk

Very low. Pure refactor of a validation helper, no public API change.