jwalk silently swallows errors on read_dir, no guidance in docs

[untitaker]

The du example swallows all errors when reading a directory, and it shows under massive parallelism as jwalk tends to run out of filedescriptors. Steps to reproduce:

1. edit the example to have parallelism 500 instead of 4
2. set ulimit -n 5 -- this step is sometimes not required, but may be required on systems with really large fd limit.
3. launch cargo run --example du on a large folder repeatedly and observe indeterministic bytecount without any printed errors

Here is a fix for that:

diff --git a/examples/du.rs b/examples/du.rs
index 6e51703..b62dbe8 100644
--- a/examples/du.rs
+++ b/examples/du.rs
@@ -23,6 +23,10 @@ fn main() {
     {
         match dir_entry_result {
             Ok(dir_entry) => {
+                if let Some(ref err) = dir_entry.read_children_error {
+                    panic!("error: {}", err);
+                }
+
                 if let Some(len) = &dir_entry.client_state {
                     eprintln!("counting {:?}", dir_entry.path());
                     total += len;

This is a duplicate of #4 -- I don't think this is good API at all, there's a user in that issue confused by it and jwalk's own examples get its usage wrong.

Additionally, if I try to observe read_children_error in process_read_dir, it is always None regardless of whether jwalk ran into ulimits.