Scraper: increase HTTP timeout and set a User-Agent header

[cuihtlauac]

Summary

The planet scraper (via the river library) has two issues causing avoidable feed fetch failures:

1. HTTP timeout is too short (3 seconds)

The timeout in river's lib/http.ml is hardcoded to 3 seconds. Several feeds that are perfectly reachable consistently time out in CI:

• https://ocamlpro.com/blog/feed — works (200) but slow
• https://mirage.io/feed.xml — works (200) but slow
• https://hannes.robur.coop/atom — intermittent
• https://blog.robur.coop/feed.xml — intermittent
• https://jon.recoil.org/atom.xml — intermittent

Proposed fix: increase the timeout to 10 seconds.

2. No User-Agent header

The scraper sends HTTP requests without a User-Agent header (cohttp default). Some sites behind Cloudflare or similar CDNs reject requests without a recognized user agent:

• https://priver.dev/tags/ocaml/index.xml — returns 403 Forbidden from CI, but 200 with a browser user agent

Proposed fix: set a common browser User-Agent header on each request. Cohttp_lwt_unix.Client.get accepts an optional ~headers parameter.

How river is managed

river is pinned to a specific commit via ocamlorg.opam.template:

pin-depends: [
  ["river.dev" "git+https://github.com/aantron/river#476dc945a908a69548bddd267f143a3e5d9c8a1a"]
]

This is a fork of kayceesrk/river. To apply the fixes:

1. Submit a PR to aantron/river (or kayceesrk/river) with the timeout and User-Agent changes
2. Update the pin hash in ocamlorg.opam.template to the new commit

Context

• Video scraper: YouTube RSS feeds returning 404 (platform-wide outage) #3512 tracks the YouTube RSS feed platform-wide outage
• Update planet sources: disable dead feeds, add Rocq Prover #3514 disables dead feeds and adds the Rocq Prover source