Drop the drain check from upgrade — agent death is routine

Agents die all the time; the pipeline's retry machinery re-queues
their work. Skipping busy sleds just created version skew and manual
retries for no real protection. `upgrade all` now sweeps every sled
unconditionally.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019fHdm92yjvguPi2LiXfLB9

server/src/chat/transport/matrix/project_rebuild.rs

@@ -343,7 +343,7 @@ async fn wait_for_drain(container_name: &str, timeout_secs: u64) -> Option<Strin
 ///
 /// Uses `docker exec <name> pgrep -f claude` — exits 0 with PID list when found,
 /// exits 1 when no matches (treated as 0 active processes).
-pub(crate) async fn count_active_claude_processes(container_name: &str) -> Result<usize, String> {
+async fn count_active_claude_processes(container_name: &str) -> Result<usize, String> {
     let out = tokio::process::Command::new("docker")
         .args(["exec", container_name, "pgrep", "-f", "claude"])
         .output()

server/src/chat/transport/matrix/sled_upgrade.rs

@@ -7,8 +7,9 @@
 //!
 //! The binary comes from the gateway's own artifact store
 //! (`~/.huskies/artifacts/`, published by the `release` command) — sleds never
-//! download from anywhere but their gateway.  Sleds with active agent
-//! processes are skipped so an upgrade never kills in-flight work.
+//! download from anywhere but their gateway.  Agents running in a sled are
+//! killed by the restart; the pipeline's retry machinery picks the work up
+//! again, same as any other agent death.
 //!
 //! The gateway orchestrates each upgrade in four phases, streaming a marker to
 //! the chat room at each step:
@@ -152,7 +153,6 @@ fn resolve_artifact_source(gateway_port: Option<u16>) -> Result<(String, Option<
 
 /// Upgrade every registered sled in sequence, streaming per-sled phase markers.
 ///
-/// Sleds with active agent processes are skipped (reported, not failed).
 /// Returns a summary listing the outcome for each sled.
 pub async fn handle_upgrade_all<F, Fut>(
     projects_store: &Arc<RwLock<BTreeMap<String, ProjectEntry>>>,
@@ -199,8 +199,8 @@ where
 /// begins.  On any failure, an error message is returned and the previous
 /// binary remains active on the sled.
 ///
-/// Sleds with active agent processes are skipped so an upgrade never kills
-/// in-flight work.
+/// Agents running in the sled are killed by the restart; the pipeline's
+/// retry machinery re-queues their work.
 pub async fn handle_sled_upgrade<F, Fut>(
     project: &str,
     projects_store: &Arc<RwLock<BTreeMap<String, ProjectEntry>>>,
@@ -236,17 +236,6 @@ where
         Err(e) => return e,
     };
 
-    // ── Drain check ──────────────────────────────────────────────────────────
-    // Never kill in-flight agent work; the caller can retry once idle.
-    let container_name = format!("huskies-{project}");
-    if let Ok(n) = super::project_rebuild::count_active_claude_processes(&container_name).await
-        && n > 0
-    {
-        return format!(
-            "skipped — {n} active agent process(es) in `{container_name}`. Retry when idle."
-        );
-    }
-
     run_sled_upgrade(project, &sled_url, &source_url, expected_hash, send_phase).await
 }