Consider creating allocID by timestamps

## Feature Request

### Describe your feature request related problem
When using `pd-recover` to recover a cluster, it is hard to find the max allocated ID from the old cluster.

### Describe the feature you'd like
When we allocate id, we can encode timestamp to the ID. For example, use high bits to store timestamp.

### Describe alternatives you've considered
We can also consider using TSO as Alloc ID directly.

### Teachability, Documentation, Adoption, Migration Strategy
Note that long IDs are hard to read and debug. We may need to make it configurable and disable by default.

/cc @tshqin