feat: add durable schedule management and multi-node crash recovery

internal/storage/mysql/migrations/000006_schedule_runtime.up.sql

@@ -0,0 +1,18 @@
+-- 旧版本只定义了结构，没有公开创建计划的入口。未绑定队列的历史行不启用扫描。
+ALTER TABLE pulse_schedules
+ ADD COLUMN queue_id BINARY(16) NULL COMMENT '计划终身绑定队列；旧结构人工行为空，不允许执行',
+ ADD COLUMN processing_version BIGINT UNSIGNED NOT NULL DEFAULT 1 COMMENT '当前处理积压的不可变版本，不因发布新版本丢弃旧进度',
+ ADD COLUMN check_at DATETIME(6) NULL COMMENT '下一可扫描时间，UTC；补跑批次之间持久化至少一秒间隔，为空表示没有扫描工作',
+ ADD CONSTRAINT ck_schedule_cursor CHECK (processing_version>0 AND processing_version<=current_version),
+ ADD CONSTRAINT ck_schedule_enabled_scan CHECK (queue_id IS NOT NULL OR check_at IS NULL),
+ ADD CONSTRAINT fk_schedule_queue FOREIGN KEY(namespace_id,queue_id) REFERENCES pulse_queues(namespace_id,id) ON DELETE RESTRICT ON UPDATE RESTRICT,
+ ADD KEY ix_schedule_scan(paused,check_at,namespace_id,id),
+ ADD KEY ix_schedule_list(namespace_id,queue_id,created_at,id);
+ALTER TABLE pulse_schedule_versions
+ ADD COLUMN catchup_window_ms BIGINT UNSIGNED NOT NULL DEFAULT 86400000 COMMENT '未处理触发点最大保留时长，单位毫秒，过旧范围记录跳过',
+ ADD COLUMN grace_ms BIGINT UNSIGNED NOT NULL DEFAULT 5000 COMMENT '正常派发允许延迟的毫秒数，超过后按漏跑策略处理',
+ ADD CONSTRAINT ck_schedule_window CHECK(catchup_window_ms BETWEEN 1000 AND 2592000000 AND grace_ms BETWEEN 1000 AND 60000 AND grace_ms<=catchup_window_ms);
+ALTER TABLE pulse_schedule_occurrences
+ MODIFY COLUMN covered_until DATETIME(6) NULL COMMENT '合并或跳过范围的排除边界，范围为原计划时间至该边界，不包含边界',
+ ADD CONSTRAINT ck_occurrence_range CHECK(covered_until IS NULL OR covered_until>scheduled_at);
+ALTER TABLE pulse_tasks ADD KEY ix_task_owner_state(namespace_id,owner_kind,owner_id,state,id);