feat(mail): deliver email one-time codes over SMTP and add the setup email screen

Felis never actually sent mail: OTP codes for onboarding, email login and
op-login were only written to the felis-api log behind a "demo has no SMTP"
limitation, and the Settings/SMTP flow those comments promised was never
built. Combined with the bootstrap Owner's address being recorded unverified
(87279a1), op-login start always took the anti-enumeration neutral branch and
minted a fake request_id, so the in-game approve inevitably answered "No
pending operator sign-in with that code".

Give the codes a real delivery path, configured in felis.toml rather than a
web settings page so config keeps a single source of truth:

- config: new [smtp] table (host, port defaulting to 587, from, username,
  password_ref). Validation requires a plausible from address and a sane
  port; the password itself never enters the config file.
- internal/mail (new): stdlib net/smtp mailer implementing the api.OTPMailer
  seam. Port 465 dials implicit TLS, other ports upgrade via STARTTLS when
  advertised; AUTH only when a username is configured (PlainAuth itself
  refuses plaintext, so the password cannot leak to a TLS-less relay).
  Ping() proves reachability and credentials without sending mail. The
  message shape (CRLF, Q-encoded bilingual subject) is pinned by test.
- platform: felis-smtp Secret constants and an optional FELIS_SMTP_PASSWORD
  env var on the felis-api Deployment, mirroring felis-uploads-s3.
- cmd/felis api: construct the real mailer when [smtp] is configured; keep
  the log fallback otherwise and say so at startup. Warn when a username is
  set but the credentials env is empty.
- setup TUI: "e" on the summary/status screen opens the email form (host,
  port, from, optional auth). Apply order: Ping preflight, [smtp] into both
  host and pod config files, felis-smtp Secret piped to kubectl via stdin,
  config Secret, felis-api rollout. A failed preflight leaves the install
  untouched. SMTP is deliberately not a wizard rail step: first-run stays
  mail-less by design, and the passkey minted at onboarding is the pre-SMTP
  owner credential.

Also make PGRepo.UserByEmail match case-insensitively (lower(email) =
lower($1)), honoring the interface contract and the users_verified_email_
unique partial index; the fake repo already matched with EqualFold.

Existing installs need the felis-api Deployment manifest re-applied (e.g. a
bootstrap re-run) before the new env var exists; a rollout restart alone
cannot add it.

cmd/felis/tui_root_test.go

@@ -223,6 +223,32 @@ func TestRootReconfigureStorageReEntersChooser(t *testing.T) {
 	}
 }
 
+// TestRootReconfigureSMTP locks the post-install "configure email" path: from
+// the re-run status screen it opens the SMTP form, and finishing it lands back
+// on the status screen (not the first-run summary, which would drop the
+// alreadySetUp framing).
+func TestRootReconfigureSMTP(t *testing.T) {
+	m := newTestRoot(true, consoleModeSetup, "")
+	m = drive(t, m, preflightDoneMsg{})
+	if _, ok := m.screen.(*summaryModel); !ok {
+		t.Fatalf("re-run after preflight, screen = %T, want *summaryModel", m.screen)
+	}
+
+	m = drive(t, m, reconfigureSMTPMsg{})
+	if _, ok := m.screen.(*smtpModel); !ok {
+		t.Fatalf("reconfigure-smtp screen = %T, want *smtpModel", m.screen)
+	}
+
+	m = drive(t, m, smtpResultMsg{configured: true, detail: "smtp.example.net:587  ·  from [email protected]"})
+	sum, ok := m.screen.(*summaryModel)
+	if !ok {
+		t.Fatalf("after reconfigure-smtp, screen = %T, want *summaryModel", m.screen)
+	}
+	if !sum.alreadySetUp {
+		t.Fatalf("after reconfigure-smtp, summary should still be the alreadySetUp status screen")
+	}
+}
+
 func TestRootRerunLandsOnStatus(t *testing.T) {
 	// adminExists at start of a setup run = re-run: preflight should skip straight
 	// to the "manage in panel" status screen, never touching owner/connect.