watchpr: fix watch mode hanging when a merged PR's head commit is gone

FetchState fetched the PR (merged=true) but then failed the whole state
fetch when CommitStatus 404'd for a head commit that no longer existed
(the branch was deleted after a squash/rebase merge). The merge signal was
discarded, so the watch loop treated every post-merge poll as a transient
error and never exited -- the 37-minute hang seen in production.

- Add a typed APIError carrying the HTTP status so callers can detect a 404
  without parsing error strings.
- FetchState now tolerates a 404 from CommitStatus (commit gone => no CI
  status) and returns the authoritative merged/closed PR state.
- Regression tests: FetchState survives a 404 status; the full watch loop,
  driven through a real client, detects a merge whose head commit is gone
  (both fail/hang before the fix).

internal/agent/watch_test.go

@@ -2,6 +2,10 @@ package agent
 
 import (
 	"errors"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"sync/atomic"
 	"testing"
 	"time"
 )
@@ -273,6 +277,78 @@ func TestWatchBaselineErrorAborts(t *testing.T) {
 	}
 }
 
+// The production hang, end to end: a watched PR stays open across several polls,
+// then is squash-merged and its branch deleted, so the commit-status endpoint
+// 404s. Driven through a real *GiteaClient, the watch loop must still detect the
+// merge on the poll it happens. Before the fix, FetchState returned an error on
+// that poll (the 404 masked the merge), so the loop reported only poll errors
+// and never exited -- exactly the 37-minute hang seen in production.
+func TestWatchDetectsMergeWhenCommitGone(t *testing.T) {
+	const sha = "cafebabecafebabe"
+	var polls atomic.Int32 // number of PR fetches so far
+
+	mux := http.NewServeMux()
+	mux.HandleFunc("/api/v1/repos/unkin/repo/pulls/7", func(w http.ResponseWriter, r *http.Request) {
+		n := polls.Add(1)
+		if n >= 4 { // baseline + two unchanged polls, then merged
+			_, _ = fmt.Fprintf(w, `{"number":7,"state":"closed","merged":true,"mergeable":true,"head":{"sha":%q}}`, sha)
+			return
+		}
+		_, _ = fmt.Fprintf(w, `{"number":7,"state":"open","merged":false,"mergeable":true,"head":{"sha":%q}}`, sha)
+	})
+	mux.HandleFunc("/api/v1/repos/unkin/repo/commits/"+sha+"/status", func(w http.ResponseWriter, r *http.Request) {
+		if polls.Load() >= 4 { // branch deleted post-merge: commit is gone
+			w.WriteHeader(http.StatusNotFound)
+			_, _ = fmt.Fprint(w, `{"message":"not found"}`)
+			return
+		}
+		_, _ = fmt.Fprint(w, `{"state":"success"}`)
+	})
+	mux.HandleFunc("/api/v1/repos/unkin/repo/issues/7/comments", func(w http.ResponseWriter, r *http.Request) {
+		_, _ = fmt.Fprint(w, `[]`)
+	})
+	srv := httptest.NewServer(mux)
+	defer srv.Close()
+
+	c := &GiteaClient{BaseURL: srv.URL, Token: "t", HTTP: srv.Client()}
+	ref := PRRef{Owner: "unkin", Repo: "repo", Number: 7}
+
+	// A real ticker so the loop advances on its own; a hang (the bug) is caught
+	// by the timeout below instead of blocking the suite.
+	tk := time.NewTicker(5 * time.Millisecond)
+	defer tk.Stop()
+	var pollErr atomic.Pointer[error]
+	type outcome struct {
+		res WatchResult
+		err error
+	}
+	done := make(chan outcome, 1)
+	go func() {
+		res, err := Watch(c, []PRRef{ref}, "unkin-agent", tk.C, nil,
+			func(_ PRRef, e error) { pollErr.Store(&e) })
+		done <- outcome{res, err}
+	}()
+
+	select {
+	case o := <-done:
+		if o.err != nil {
+			t.Fatalf("Watch: %v", o.err)
+		}
+		if o.res.Reason != "PR merged" {
+			t.Errorf("reason = %q, want %q", o.res.Reason, "PR merged")
+		}
+		if p := pollErr.Load(); p != nil {
+			t.Errorf("no poll error expected once a 404 status is tolerated, got: %v", *p)
+		}
+	case <-time.After(3 * time.Second):
+		var got error
+		if p := pollErr.Load(); p != nil {
+			got = *p
+		}
+		t.Fatalf("Watch hung: a merge with a gone head commit was never detected (last poll error: %v)", got)
+	}
+}
+
 func TestCountNonAgentComments(t *testing.T) {
 	comments := []Comment{
 		{User: User{Login: "unkin-agent"}},