fix(ecstore): classify decommission stage failures by type, not by message (#6269)

T2 of backlog#1827. `data_movement_stage_error` flattened every stage failure into `Error::other(format!(...))`, discarding the typed error. The cost was visible in tree: `is_decommission_target_capacity_error` had to match rendered text —

    let message = err.to_string();
    message.contains(&disk_full) || message.contains(&storage_full)

— to notice that the destination pool had filled up, and `is_decommission_copy_cleanup_safe_error` could not see a not-found that surfaced from inside a stage at all.

The wrapper now carries what it wrapped. `DataMovementStageError` renders the same string and returns the original through `source()`; `Error::other` boxes it through `std::io::Error`, so `data_movement_stage_source` recovers it by downcast. Both classifiers unwrap before matching, keeping their substring paths for errors that arrive through some other wrapper.

The rendered message is unchanged, which a test now pins against the exact string the old `format!` produced rather than against a `contains`. Three more cover the round trip for `DiskFull`, `StorageFull`, `FileNotFound` and `SlowDown`, that unrelated errors are not mistaken for stage wrappers, and — the case the issue names — that a not-found surfacing from inside a stage is judged cleanup-safe by the decommission loop exactly as a direct one is.

Refs backlog#1827

crates/ecstore/src/core/pools.rs

@@ -1041,7 +1041,13 @@ fn should_count_decommission_version_complete(ignore: bool, cleanup_ignored: boo
 fn is_decommission_copy_cleanup_safe_error(err: &Error) -> bool {
     // DataMovementOverwriteErr only means source and destination pool resolved to
     // the same pool. Without a target equivalence check it is not cleanup-safe.
-    is_err_object_not_found(err) || is_err_version_not_found(err)
+    if is_err_object_not_found(err) || is_err_version_not_found(err) {
+        return true;
+    }
+
+    // A not-found surfacing from inside a data-movement stage is the same
+    // condition once the wrapper is unwrapped (backlog#1827 T2).
+    crate::data_movement::data_movement_stage_source(err).is_some_and(is_decommission_copy_cleanup_safe_error)
 }
 
 fn is_decommission_target_capacity_error(err: &Error) -> bool {
@@ -1049,6 +1055,13 @@ fn is_decommission_target_capacity_error(err: &Error) -> bool {
         return true;
     }
 
+    // A stage failure keeps the error it wrapped, so classify by type rather
+    // than by the rendered message (backlog#1827 T2). The substring fallback
+    // stays for errors that reached here through some other wrapper.
+    if let Some(source) = crate::data_movement::data_movement_stage_source(err) {
+        return is_decommission_target_capacity_error(source);
+    }
+
     let message = err.to_string();
     let disk_full = Error::DiskFull.to_string();
     let storage_full = Error::StorageFull.to_string();
@@ -4427,6 +4440,36 @@ mod tests {
         assert!(is_decommission_target_capacity_error(&Error::StorageFull));
     }
 
+    /// The decommission loop classifies errors that came back through a
+    /// data-movement stage wrapper. Before backlog#1827 T2 the wrapper flattened
+    /// everything into `Error::other(String)`, so these two classifiers had to
+    /// match on rendered text; now the wrapped error is recoverable by type.
+    #[test]
+    fn decommission_classifiers_see_through_a_stage_wrapper() {
+        let wrap = |inner: Error| {
+            crate::data_movement::data_movement_stage_error_for_test(
+                "decommission_object",
+                "put_object",
+                "bucket-a",
+                "object-a",
+                inner,
+            )
+        };
+
+        // Capacity: the target pool filling up must still stop the loop.
+        assert!(is_decommission_target_capacity_error(&wrap(Error::DiskFull)));
+        assert!(is_decommission_target_capacity_error(&wrap(Error::StorageFull)));
+        assert!(!is_decommission_target_capacity_error(&wrap(Error::SlowDown)));
+
+        // Cleanup safety: a not-found surfacing from inside a stage is the same
+        // condition as one surfacing directly, so the source entry stays
+        // eligible for cleanup.
+        let not_found = Error::ObjectNotFound("bucket-a".to_string(), "object-a".to_string());
+        assert!(is_decommission_copy_cleanup_safe_error(&not_found));
+        assert!(is_decommission_copy_cleanup_safe_error(&wrap(not_found)));
+        assert!(!is_decommission_copy_cleanup_safe_error(&wrap(Error::SlowDown)));
+    }
+
     #[test]
     fn decommission_target_capacity_error_accepts_wrapped_capacity_errors() {
         let disk_full = Error::other(format!("decommission_object: put_object failed for bucket/object: {}", Error::DiskFull));