builtin/add: enable fscache around repo_read_index_preload (#899)

Trace2 + GIT_TRACE_FSCACHE evidence on Windows ARM64 (Snapdragon X
Elite, ReFS Dev Drive) shows that the heaviest lstat-bound work in `git
add` happens inside repo_read_index_preload(), which currently runs
*before* enable_fscache() is called. Moving the enable up so the preload
phase is wrapped lets the existing batched NtQueryDirectoryFile cache
cover the bulk of the lstat traffic. This patch gave me a ~30%
performance improvement on a large git repo with a batched add.

Also at the end of cmd_add(): the cleanup site called enable_fscache(0)
again instead of disable_fscache(), leaking the refcount.

Author: dscho

builtin/add.c

@@ -496,13 +496,13 @@ int cmd_add(int argc,
 		 (!(addremove || take_worktree_changes)
 		  ? ADD_CACHE_IGNORE_REMOVAL : 0));
 
+	enable_fscache(0);
 	if (repo_read_index_preload(repo, &pathspec, 0) < 0)
 		die(_("index file corrupt"));
 
 	die_in_unpopulated_submodule(repo->index, prefix);
 	die_path_inside_submodule(repo->index, &pathspec);
 
-	enable_fscache(0);
 	/* We do not really re-read the index but update the up-to-date flags */
 	preload_index(repo->index, &pathspec, 0);
 
@@ -621,6 +621,6 @@ int cmd_add(int argc,
 	free(ps_matched);
 	dir_clear(&dir);
 	clear_pathspec(&pathspec);
-	enable_fscache(0);
+	disable_fscache();
 	return exit_status;
 }