Move NFS-cache busting to where it's really needed

The NFS directory cache is stored on a per-client basis, so the lexbox
backend pod isn't the place we need to run `ls` to invalidate the NFS
cache. The hgweb pod is where the `ls` needs to happen.

backend/LexBoxApi/Services/HgService.cs

@@ -104,8 +104,6 @@ public async Task ResetRepo(string code)
         await SoftDeleteRepo(code, $"{FileUtils.ToTimestamp(DateTimeOffset.UtcNow)}__reset");
         //we must init the repo as uploading a zip is optional
         tmpRepo.MoveTo(PrefixRepoFilePath(code));
-        // Force NFS to refresh its directory cache.
-        Process.Start("ls", [PrefixRepoFilePath(code)]);
     }
 
     public async Task FinishReset(string code, Stream zipFile)
@@ -139,8 +137,6 @@ await Task.Run(() =>
         // Now we're ready to move the new repo into place, replacing the old one
         await DeleteRepo(code);
         tempRepo.MoveTo(PrefixRepoFilePath(code));
-        // Force NFS to refresh its directory cache.
-        Process.Start("ls", [PrefixRepoFilePath(code)]);
     }
 
     /// <summary>

hgweb/command-runner.sh

@@ -35,6 +35,8 @@ echo ""
 
 # Run the hg command, simply output to stdout
 first_char=$(echo $project_code | cut -c1)
+# Ensure NFS cache is refreshed in case project repo changed in another pod (e.g., project reset)
+ls /var/hg/repos/$first_char/$project_code > /dev/null  # Don't need output; this is enough to refresh NFS dir cache
 cd /var/hg/repos/$first_char/$project_code
 case $command_name in