converter.sh: Don't parallelise write_hash (#17742)

write_hash appends to scratch_files/hashes.csv without any form of
locking, but is performed in parallel. Additionally there is no
synchronisation to wait for all write_hash invocations to have completed
before reading the CSV. This results in non-deterministic breakage and
missing hashes in the resulting files.

Remove parallelism here to fix the breakage.

Note, just calling wait_for_jobs after the loop reduces the occurrence
of problems, but is insufficient since multiple appends do occasionally
race and hashes may still be missing.

converter.sh

@@ -296,7 +296,7 @@ function write_hash () {
 }
 
 while IFS=, read -r gid predicate path
-    do write_hash "${predicate}" "${path}" "${gid}" "scratch_files/hashes.csv" &
+    do write_hash "${predicate}" "${path}" "${gid}" "scratch_files/hashes.csv"
 done < scratch_files/paths.csv > /dev/null
 
 jq -cR 'split(",")' scratch_files/hashes.csv | jq -s 'map({(.[0]): [.[1], .[2]]}) | add' > scratch_files/hashmap.json