sbin deploy can't set HADOOP_CONF_DIR for taskmanager

and spark extra conf, e.g. hdfs-site.xml, needs to add in spark/conf before deploy. -- HADOOP_CONF_DIR设置为openmldb/conf应该可以？
但taskmanager读取需要start.sh时有环境，sbin不会传去host里
just deploy conf for fix config options

More

stat-all.sh没有recover工具时最好failed，除非force执行，不然用户很容易重启后表有问题，但当时无法察觉
start-all.sh里的recovering没有进度，前台也没法看，没有多开shell时很麻烦
recovering本来就是在tablet、nameserver都启动后的操作，在线集群可以独立recovering，不做不必要的stop

yarn job如果日志出现connect 0.0.0.0:8032，说明yarn-site.xml没有正确load。使用了默认yarn地址。

Look at sbin deploy scripts to see how environment variables like HADOOP_CONF_DIR are passed to taskmanager. Examine start.sh and related host deployment logic. The fix likely involves modifying deployment scripts to propagate configuration. Check spark/conf handling for hdfs-site.xml. Also review stat-all.sh and start-all.sh for recovery behavior improvements mentioned in the issue.

由索引模型根据 Issue 内容生成。