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/jobs/registry_shared.go

@@ -0,0 +1,115 @@
+package jobs
+
+import (
+	"context"
+	"errors"
+
+	"evobgp/internal/store"
+)
+
+// FindIdempotentContext looks up a retained logical operation without inserting.
+func (r *Registry) FindIdempotentContext(ctx context.Context, tenant, key string) (*Job, error) {
+	if r.queue != nil {
+		c, err := r.queue.FindJobByIdempotencyKey(ctx, tenant, key)
+		if err != nil {
+			return nil, err
+		}
+		j := jobFromRecord(c)
+		j.requestHash = c.RequestHash
+		return j, nil
+	}
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	j := r.byIdempo[idempoKey{tenant: tenant, key: key}]
+	if j == nil {
+		return nil, store.ErrJobNotFound
+	}
+	return j, nil
+}
+
+// ReplayContext validates an immutable request before expensive live resolution.
+func (r *Registry) ReplayContext(ctx context.Context, in store.JobEnqueue) (*Job, bool, error) {
+	if in.IdempotencyKey == nil || *in.IdempotencyKey == "" {
+		return nil, false, nil
+	}
+	j, err := r.FindIdempotentContext(ctx, in.TenantID, *in.IdempotencyKey)
+	if errors.Is(err, store.ErrJobNotFound) {
+		return nil, false, nil
+	}
+	if err != nil {
+		return nil, false, err
+	}
+	h, err := store.JobRequestHash(in)
+	if err != nil {
+		return nil, false, err
+	}
+	if j.requestHash != h {
+		return nil, false, store.ErrJobConflict
+	}
+	return j, true, nil
+}
+
+// GetContext reads a job from the shared store when durability is enabled.
+func (r *Registry) GetContext(ctx context.Context, tenant, id string) (*Job, error) {
+	if r.queue == nil {
+		return r.Get(tenant, id)
+	}
+	c, err := r.queue.GetJob(ctx, tenant, id)
+	if errors.Is(err, store.ErrJobNotFound) {
+		return nil, ErrNotFound
+	}
+	if err != nil {
+		return nil, err
+	}
+	return jobFromRecord(c), nil
+}
+
+// ListContext returns a bounded tenant page from the authoritative queue.
+func (r *Registry) ListContext(ctx context.Context, tenant, status, kind, cursor string, limit int) ([]*Job, string, bool, error) {
+	if r.queue == nil {
+		j, n, m := r.List(tenant, status, kind, cursor, limit)
+		return j, n, m, nil
+	}
+	c, next, more, err := r.queue.ListJobs(ctx, tenant, status, kind, cursor, limit)
+	if err != nil {
+		return nil, "", false, err
+	}
+	out := make([]*Job, 0, len(c))
+	for _, v := range c {
+		out = append(out, jobFromRecord(v))
+	}
+	return out, next, more, nil
+}
+
+// RequestCancelContext persists a cancel request visible to any worker replica.
+func (r *Registry) RequestCancelContext(ctx context.Context, tenant, id string) (*Job, error) {
+	if r.queue == nil {
+		return r.RequestCancel(tenant, id)
+	}
+	c, err := r.queue.CancelJob(ctx, tenant, id)
+	if errors.Is(err, store.ErrJobNotFound) {
+		return nil, ErrNotFound
+	}
+	if err != nil {
+		return nil, err
+	}
+	r.mu.RLock()
+	local := r.byID[id]
+	r.mu.RUnlock()
+	if local != nil {
+		local.RequestCancel()
+	}
+	return jobFromRecord(c), nil
+}
+
+// RetryContext schedules an explicit retry of a failed shared operation.
+func (r *Registry) RetryContext(ctx context.Context, tenant, id string) (*Job, error) {
+	if r.queue == nil {
+		return nil, store.ErrJobConflict
+	}
+	c, err := r.queue.RetryJob(ctx, tenant, id)
+	if err != nil {
+		return nil, err
+	}
+	return jobFromRecord(c), nil
+}