Work in progress!
fixes coder/internal#455

This PR moves incrementing the workspace build number into the insert workspace build query so the operation is atomic.
I'm not exactly sure if this is the right fix yet.

Theory:
You can hit a race condition where the lifecycle executor takes a lock on the workspace, but it's only scoped to prevent multiple lifecycle executors from operating on the workspace at the same time:
ok, err := tx.TryAcquireLock(e.ctx, database.GenLockID(fmt.Sprintf("lifecycle-executor:%s", wsID)))
A manual action (like clicking "start workspace" or perhaps an API call in a concurrent build) can compete with a lifecycle executor.

I need to dig into how concurrent builds work to actually verify what's going on here.

Edit: the fix will probably involve moving the build_number field over to a serial type, it seems I just moved the race condition to the database