uphold latency requirement when submitting blocking operations

If a user submits a blocking operation from a latency-sensitive queue,
the blocking thread should unconditionally wake the executor to trigger
a preemption. This is the same principle that applies to foreign
wakes.

glommio/src/sys/blocking.rs

@@ -108,6 +108,7 @@ impl TryFrom<BlockingThreadResult> for std::io::Result<usize> {
 
 pub(super) struct BlockingThreadReq {
     op: BlockingThreadOp,
+    latency_sensitive: bool,
     id: u64,
 }
 
@@ -137,7 +138,7 @@ impl BlockingThread {
                 if tx.send(resp).is_err() {
                     panic!("failed to send response");
                 }
-                reactor_sleep_notifier.notify(false);
+                reactor_sleep_notifier.notify(el.latency_sensitive);
             }
         }))
     }
@@ -191,7 +192,14 @@ impl BlockingThreadPool {
         let id = self.requests.get();
         self.requests.set(id.overflowing_add(1).0);
 
-        let req = BlockingThreadReq { op, id };
+        let req = BlockingThreadReq {
+            op,
+            id,
+            latency_sensitive: matches!(
+                source.borrow().io_requirements.latency_req,
+                crate::Latency::Matters(_)
+            ),
+        };
 
         self.tx.send(req).map_err(|_| {
             io::Error::new(