singlefile: force --window-size=1920,1080 for viewport-dependent CSS preservation

Add --window-size=1920,1080 to the Chromium flags passed via --browser-args.
This makes the existing --remove-unused-styles=false and
--remove-alternative-medias=false effective for real @media rules
and responsive styles (headless default is small).

Also document in ARCHIVR_CHROME_ARGS that users can override by
supplying their own --window-size in the env var.

crates/archivr-core/src/downloader/singlefile.rs

@@ -72,9 +72,16 @@ fn save_with(
     // then append any extra flags from ARCHIVR_CHROME_ARGS (space-separated).
     // Docker containers running as root need "--no-sandbox" here because
     // Chromium refuses to start as root without it.
+    //
+    // --window-size is set to a realistic desktop viewport so that
+    // --remove-alternative-medias=false and --remove-unused-styles=false
+    // actually preserve responsive @media rules and styles that only match
+    // at normal screen widths (headless Chromium defaults to a small viewport
+    // that would otherwise defeat the preservation flags).
     let mut chrome_flags = vec![
         "--disable-web-security".to_string(),
         format!("--user-data-dir={}", chrome_data_dir.display()),
+        "--window-size=1920,1080".to_string(),
     ];
     if let Ok(extra) = std::env::var("ARCHIVR_CHROME_ARGS") {
         chrome_flags.extend(extra.split_whitespace().filter(|s| !s.is_empty()).map(str::to_string));

docs/README.md

@@ -314,6 +314,9 @@ dependencies (Chromium, Node.js, Python) land in the final layer.
   - Space-separated extra flags appended to Chromium's `--browser-args`. The Docker
     image sets this to `--no-sandbox` because Chromium refuses to run as root without
     it. Leave unset when running natively (Nix, Linux desktop).
+    A `--window-size=1920,1080` is always passed to provide a realistic desktop
+    viewport (so responsive @media rules and styles are evaluated and preserved
+    correctly). Supply your own `--window-size=...` here to override.
 - `ARCHIVR_TWITTER_CREDENTIALS_FILE`
   - Required for tweet/thread scraping inputs such as `tweet:ID` and `x:thread:ID`.
   - Must point to a cookies file for the vendored scraper.