Merge origin/main into fix/reconciliation-sweep-flakes

The fake serves one request per socket and drops it, but its responses
carried no Connection: close, so the client kept the dead socket in its
pool. With retries disabled (as in production) a reused dead connection
failed the restore, and sweep_restores_tigris_cache_miss_and_skips_missing
skipped its repo with 'repo acquire failed' on a loaded runner.

crates/gitlawb-node/src/reconciliation.rs

@@ -6440,6 +6440,13 @@ mod tests {
     /// Write one canned HTTP response on an accepted fake-S3 socket.
     /// A free function (not a closure) so the `&mut` socket borrow does
     /// not leak into a returned future's lifetime.
+    ///
+    /// The fake serves one request per accepted socket and drops it, so each
+    /// response says `Connection: close`. Without it the client keeps the
+    /// socket in its pool after the server has closed it, and the next request
+    /// can reuse a dead connection. The client has retries disabled, as in
+    /// production, so `sweep_restores_tigris_cache_miss_and_skips_missing`
+    /// failed with "repo acquire failed" on a loaded runner.
     async fn s3_respond(
         sock: &mut tokio::net::TcpStream,
         status: &str,
@@ -6448,7 +6455,7 @@ mod tests {
     ) {
         use tokio::io::AsyncWriteExt;
         let head = format!(
-            "HTTP/1.1 {status}\r\nContent-Length: {}\r\n{extra}\r\n",
+            "HTTP/1.1 {status}\r\nContent-Length: {}\r\nConnection: close\r\n{extra}\r\n",
             payload.len()
         );
         let _ = sock.write_all(head.as_bytes()).await;