internal/dave: improve snapshot archive storage logic

[joshuasing]

Currently, Dave uses a Repository to store snapshots. Internally, a Repository looks like this:

// Repository is a storage implementation which stores snapshots.
type Repository interface {
	Metadata(ctx context.Context) (*RepositoryMeta, error)
	MetadataUpdate(ctx context.Context, meta *RepositoryMeta) error

	SnapshotAdd(ctx context.Context, snapshot *Snapshot) error
	SnapshotByID(ctx context.Context, id string) (*Snapshot, error)
	SnapshotList(ctx context.Context) ([]*Snapshot, error)
	SnapshotRemove(ctx context.Context, id string) error
}

A Snapshot consists of metadata and one or more archives. This means that we only add a snapshot once all archives are created.

Archives are stored on the local file system before being moved to the repository, requiring enough storage for:

1. The data being backed up (e.g. 1TB)
2. The clone of data being backed up (same as 1; 1TB)
3. All archives of the cloned data (same as 2, compressed, e.g. 0.8x)

In cases where we are making a snapshot of multiple very large directories, it may work better to upload snapshot archives individually to the repository as they are created, slightly reducing the storage required to use Dave.