[ML] Fix unit test deadlock problem (#34174)

This change fixes a potential deadlock problem in the unit
test introduced in #34117.

It also removes a piece of debug code and corrects a docs
formatting problem that were both added in that same PR.

docs/reference/ml/apis/find-file-structure.asciidoc

@@ -1284,7 +1284,7 @@ If the request does not encounter errors, you receive the following result:
 
 If you try to analyze a lot of data then the analysis will take a long time.
 If you want to limit the amount of processing your {es} cluster performs for
-a request, use the timeout query parameter. The analysis will be aborted and
+a request, use the `timeout` query parameter. The analysis will be aborted and
 an error returned when the timeout expires. For example, you can replace 20000
 lines in the previous example with 200000 and set a 1 second timeout on the
 analysis:

x-pack/plugin/ml/src/test/java/org/elasticsearch/xpack/ml/filestructurefinder/FileStructureFinderManagerTests.java

@@ -157,16 +157,16 @@ public class FileStructureFinderManagerTests extends FileStructureTestCase {
                     // Expected if timeout occurs and the input stream is closed before junk generation is complete
                 }
             });
-            junkProducer.start();
 
             try (InputStream bigInput = new PipedInputStream(generator)) {
 
+                junkProducer.start();
+
                 ElasticsearchTimeoutException e = expectThrows(ElasticsearchTimeoutException.class,
                     () -> structureFinderManager.findFileStructure(explanation, linesOfJunk - 1, bigInput, EMPTY_OVERRIDES, timeout));
 
                 assertThat(e.getMessage(), startsWith("Aborting structure analysis during ["));
                 assertThat(e.getMessage(), endsWith("] as it has taken longer than the timeout of [" + timeout + "]"));
-                explanation.add(e.getMessage());
             }
 
             // This shouldn't take anything like 10 seconds, but VMs can stall so it's best to