fix: Revert #22017 and improve block(_in_place)_on doc comment (#22031)

Author: orlp

Diff for: crates/polars-io/src/pl_async.rs

@@ -295,21 +295,20 @@ impl RuntimeManager {
         Self { rt }
     }
 
-    /// Shorthand for `tokio::task::block_in_place(|| block_on(f))`. This is a variant of `block_on`
-    /// that is safe to call from if the current thread has already entered the async runtime, or
-    /// is a rayon thread.
-    ///
-    /// # Safety
-    /// The tokio runtime flavor is multi-threaded.
+    /// Forcibly blocks this thread to evaluate the given future. This can be
+    /// dangerous and lead to deadlocks if called re-entrantly on an async
+    /// worker thread as the entire thread pool can end up blocking, leading to
+    /// a deadlock. If you want to prevent this use block_on, which will panic
+    /// if called from an async thread.
     pub fn block_in_place_on<F>(&self, future: F) -> F::Output
     where
         F: Future,
     {
         tokio::task::block_in_place(|| self.rt.block_on(future))
     }
 
-    /// Note: `block_in_place_on` should be used instead if the current thread is a rayon thread or
-    /// has already entered the async runtime.
+    /// Blocks this thread to evaluate the given future. Panics if the current
+    /// thread is an async runtime worker thread.
     pub fn block_on<F>(&self, future: F) -> F::Output
     where
         F: Future,

Diff for: crates/polars-stream/src/execute.rs

@@ -233,7 +233,7 @@ fn run_subgraph(
         if std::env::var("POLARS_TRACK_WAIT_STATS").as_deref() == Ok("1") {
             async_executor::track_task_wait_statistics(true);
         }
-        let ret = polars_io::pl_async::get_runtime().block_in_place_on(async move {
+        let ret = polars_io::pl_async::get_runtime().block_on(async move {
             for handle in join_handles {
                 handle.await?;
             }