fix: bump revision & compact when restoring

[FabianKramm]

resolves ENG-7912

With old restore method (before this patch) there is an issue where we do not bump the revision number or compact it which leads to watches getting stale as the revision number decreases.

Changes:

• Biggest changes come from dependency "modernc.org/sqlite v1.29.10", which we already have in vcluster-pro so ultimately this shouldn't add much new code. The rest is from requiring "go.etcd.io/etcd/server/v3/storage/backend", which is needed for reading the raw etcd database
• We get the latest revision before restoring from the current database and then bumping it when starting the backing store
• We also compact the latest revision after restoring the database
• New safeguard function that reverts changes being done should the restore fail

[johannesfrey]

lgtm