Use unblocked files, not components, to determine bucket sizes

Summary:
In D8758296 I changed the max_bucket_size constant to mean the number of files
in a bucket rather than the number of components. However, I did not correctly
update the logic to calculate the actual bucket size.

The logic error became clear in D13972924 where I refactored the code to use
the name `ready_components` instead of `jobs`. We are comparing files and
components, when we should be comparing files and files.

Reviewed By: nmote

Differential Revision: D13993319

fbshipit-source-id: fdc7d4d7075b6e06fb60618457248cc7dae61479

src/services/inference/merge_stream.ml

@@ -82,11 +82,11 @@ let max_bucket_size = 500
 let bucket_size stream =
   (* NB: num_workers can be zero *)
   let max_bucket_size =
-    if stream.ready_components < stream.num_workers * max_bucket_size
-    then 1 + (stream.ready_components / stream.num_workers)
+    if stream.ready_files < stream.num_workers * max_bucket_size
+    then 1 + (stream.ready_files / stream.num_workers)
     else max_bucket_size
   in
-  min max_bucket_size stream.ready_components
+  min max_bucket_size stream.ready_files
 
 let is_done stream =
   stream.blocked_components = 0