patch 9.0.1723: Fix regression in {func} argument of reduce()

Problem: Fix regression in {func} argument of reduce()
Solution: pass function name as string again

Before patch 9.0.0548, passing a string as {func} argument of reduce()
is treated as a function name, but after patch 9.0.0548 it is treated as
an expression instead, which is useless as reduce() doesn't set any v:
variables. This PR restores the behavior of {func} before that patch.

Also correct an emsg() call, as e_string_list_or_blob_required doesn't
contain format specifiers.

closes: #12824

Signed-off-by: Christian Brabandt <cb@256bit.org>
Co-authored-by: zeertzjq <zeertzjq@outlook.com>
diff --git a/src/filepath.c b/src/filepath.c
index 79d4afb..c30d9bf 100644
--- a/src/filepath.c
+++ b/src/filepath.c
@@ -1616,7 +1616,7 @@
 	argv[0].vval.v_dict = dict;
     }
 
-    if (eval_expr_typval(expr, argv, 1, NULL, &rettv) == FAIL)
+    if (eval_expr_typval(expr, FALSE, argv, 1, NULL, &rettv) == FAIL)
 	goto theend;
 
     // We want to use -1, but also true/false should be allowed.