fix(archsetup): log the real exit code from failed installs

`retry_install` was logging "error code: 0" for every retry-exhausted install (6 AUR packages on the 2026-05-16 run, 2 on the 2026-05-18 run). Root cause: `last_exit_code=$?` ran AFTER an `if eval ...; then return 0; fi` block. Bash defines an if-compound's exit status as zero when no condition tested true. With no else clause and a failing eval, that's exactly what happens, so `$?` captures 0 instead of eval's actual exit code.

Now eval runs unconditionally, `$?` lands in `last_exit_code` right after it, and the if just compares against the captured value. Same control flow, accurate codes in the summary. Verified with a small bash test: the old pattern reports `$?=0` after a failing if, the new pattern reports the real exit.

1 files changed, 7 insertions, 2 deletions

diff --git a/archsetup b/archsetup
index d044d1c..6c17232 100755
--- a/archsetup
+++ b/archsetup
@@ -493,10 +493,15 @@ retry_install() {
 
     display "task" "installing $pkg via $source"
     while [ $attempt -le $MAX_INSTALL_RETRIES ]; do
-        if eval "$cmd" >> "$logfile" 2>&1; then
+        # Capture $? from eval directly. A naked `last_exit_code=$?` after
+        # `if eval; then ...; fi` reads the if-compound's exit status, which
+        # bash defines as 0 when no condition tested true — so a failing
+        # eval gets logged as "error code: 0" downstream.
+        eval "$cmd" >> "$logfile" 2>&1
+        last_exit_code=$?
+        if [ $last_exit_code -eq 0 ]; then
             return 0
         fi
-        last_exit_code=$?
         attempt=$((attempt + 1))
         if [ $attempt -le $MAX_INSTALL_RETRIES ]; then
             display "task" "retrying $pkg (attempt $attempt/$MAX_INSTALL_RETRIES)"