[SkyPilot] Scheduling that takes into account the filesystem available in each cluster

Something like adding filesystem paths as constraints in the job launcher file: ``` ... required_filesystems: - /path/to/dir1 - /path/to/dir2 ``` so the job only gets scheduled in clusters where the required filesystems are present. Extension: Supporting different required filesystems depending on the cluster: ``` ... required_filesystems: cluster 1: - /path/to/dir1 cluster 2: - /path/to/dir2 ``` the goal is to solve for cases where the same dataset lives under different locations in different clusters: ``` cluster 1: /data/MNIST/... cluster 2: /home/$user/data/MNIST/... ```