phraser: the reminder summary is cut in runes, and silence is an error (V-620)

Three defects in internal/phraser, all of the shape "reports done when
nothing happened".

The reminder summary was cut in bytes: `len(s) > 60` and `s[:57]`, in two
copies (Stub.PhraseReminder and LLMPhraser.PhraseReminder). On Russian a
letter is two bytes, so the cut fell at about 28 letters instead of 60 and
landed inside a letter about half the time. Sendable.Summary is what
voicesink hands to piper and what the telegram sink posts, so the half rune
was spoken and sent. One rune-counting helper now, shared by both. The test
that covered this was ASCII, which is what let the arithmetic stand.

The evidence branch of PhraseQuery and the bare-prose tail of PhraseChat
both returned ("", nil) when the server answered and the model wrote no
tokens. The knowledge branch has guarded that with errEmptyResponse since it
was written; these two did not. The daemon's callers check for the empty
string and paper over it, so the visible cost was the eval, which scored a
silent model as bad phrasing rather than as a failure, and a log line that
never appeared.

Stub.PhraseReminder set no Mood. Its sibling PhraseNudge sets "neutral" and
says in a comment why: the Stub is a production fallback and owes the output
contract a value. The zero value is not one of the five moods.

No prompt and no spoken wording changed, so the phrasing eval is unmoved.

internal/phraser/phraser_test.go

@@ -5,6 +5,7 @@ import (
 	"strings"
 	"testing"
 	"time"
+	"unicode/utf8"
 
 	"github.com/kami/maven/internal/delivery"
 	"github.com/kami/maven/internal/dialogue"
@@ -185,6 +186,34 @@ func TestPhraseReminderTruncatesLongSummary(t *testing.T) {
 	}
 }
 
+// The same truncation, in the language she actually speaks. The test above is
+// ASCII, which is what let the byte arithmetic stand: `len(s) > 60` and `s[:57]`
+// cut a Russian reminder at about 28 letters instead of 60, and landed inside a
+// letter about half the time. Summary is what voicesink hands to piper and what
+// the telegram sink posts, so half a rune was spoken and sent.
+func TestPhraseReminderSummaryCountsRunesNotBytes(t *testing.T) {
+	long := "позвонить маме и забрать посылку из пункта выдачи на соседней улице до восьми вечера"
+	rd := loop.ReminderDecision{
+		Reminder: store.Reminder{Payload: `{"text":"` + long + `"}`},
+		State:    loop.State{Now: time.Now().UTC()},
+	}
+	pr, _ := NewStub().PhraseReminder(context.Background(), rd)
+	if !utf8.ValidString(pr.Summary) {
+		t.Fatalf("summary is not valid UTF-8, it was cut mid-letter: %q", pr.Summary)
+	}
+	if n := utf8.RuneCountInString(pr.Summary); n > summaryLimit {
+		t.Fatalf("summary = %d runes, want at most %d: %q", n, summaryLimit, pr.Summary)
+	}
+	// The cut must be near the limit, not near half of it. A byte count would
+	// stop at 28 letters here.
+	if n := utf8.RuneCountInString(pr.Summary); n < summaryLimit-5 {
+		t.Fatalf("summary = %d runes, cut far too early — counted in bytes? %q", n, pr.Summary)
+	}
+	if pr.Mood == "" {
+		t.Error("Mood is empty; the Stub is a production fallback and owes the contract a mood")
+	}
+}
+
 func TestPhraseReminderNonJSONPayload(t *testing.T) {
 	// a payload that isn't JSON → the phraser falls back to the raw string.
 	rd := loop.ReminderDecision{