feat(httpapi): harden control plane and speed up prefix refresh

Очередь job получила lease и fencing, телеметрия и прогресс deploy сохраняются, ревизия пишет manifest входов. Обновление префиксов не переписывает неизменённый снимок, большие наборы идут через COPY, а устаревшая ревизия не попадает в deploy. Бандл ноды кэшируется, prune и постановка job больше не держат горячий путь.

Co-authored-by: Cursor <cursoragent@cursor.com>

internal/nodedispatch/telemetry_limits_test.go

@@ -0,0 +1,80 @@
+package nodedispatch_test
+
+import (
+	"context"
+	"errors"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+)
+
+type telemetryTestBody struct {
+	io.Reader
+	read   int
+	closed bool
+}
+
+func (b *telemetryTestBody) Read(p []byte) (int, error) {
+	n, err := b.Reader.Read(p)
+	b.read += n
+	return n, err
+}
+func (b *telemetryTestBody) Close() error { b.closed = true; return nil }
+
+type telemetryErrorReader struct{}
+
+func (telemetryErrorReader) Read([]byte) (int, error) { return 0, errors.New("injected body failure") }
+
+func TestAgentTelemetryCompleteBodyLimits(t *testing.T) {
+	sp := &store.Speaker{ID: "legacy", Role: "replica", MetaJSON: store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: "agent.example.test", AgentSecret: "secret"})}
+	for _, tc := range []struct {
+		name, prefix string
+		limit        int
+		call         func(nodedispatch.Options) string
+	}{
+		{"health", `{"ok":true}`, 64 << 10, func(opts nodedispatch.Options) string {
+			return nodedispatch.FetchAgentHealth(context.Background(), sp, opts).Error
+		}},
+		{"protocols", `{"sessions":[]}`, 8 << 20, func(opts nodedispatch.Options) string {
+			return nodedispatch.FetchBirdProtocols(context.Background(), sp, opts).Error
+		}},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			for _, known := range []bool{true, false} {
+				for _, extra := range []int{0, 1} {
+					data := tc.prefix + strings.Repeat(" ", tc.limit-len(tc.prefix)+extra)
+					body := &telemetryTestBody{Reader: strings.NewReader(data)}
+					client := &http.Client{Transport: legacyRoundTripFunc(func(*http.Request) (*http.Response, error) {
+						length := int64(-1)
+						if known {
+							length = int64(len(data))
+						}
+						return &http.Response{StatusCode: 200, ContentLength: length, Body: body, Header: make(http.Header)}, nil
+					})}
+					err := tc.call(nodedispatch.Options{HTTPClient: client})
+					if (extra == 0 && err != "") || (extra == 1 && err == "") || body.read > tc.limit+1 || !body.closed {
+						t.Fatalf("known=%v extra=%d read=%d closed=%v error=%s", known, extra, body.read, body.closed, err)
+					}
+				}
+			}
+			for _, status := range []int{200, 400} {
+				reader := io.MultiReader(strings.NewReader(tc.prefix), telemetryErrorReader{})
+				if status == 400 {
+					reader = io.MultiReader(strings.NewReader(strings.Repeat("x", 16000)))
+				}
+				body := &telemetryTestBody{Reader: reader}
+				client := &http.Client{Transport: legacyRoundTripFunc(func(*http.Request) (*http.Response, error) {
+					return &http.Response{StatusCode: status, ContentLength: -1, Body: body, Header: make(http.Header)}, nil
+				})}
+				err := tc.call(nodedispatch.Options{HTTPClient: client})
+				if err == "" || !body.closed || (status == 400 && body.read > 4<<10) {
+					t.Fatalf("status=%d read=%d closed=%v error=%s", status, body.read, body.closed, err)
+				}
+			}
+		})
+	}
+}