Only create cache device placeholder when encryption is enabled

[jbaublitz]

We got a suggestion to remove the placeholder cache device in cases where the pool is not encrypted to avoid overhead. Per our discussions with @oniko, we believe that once the pool is encrypted, we should not be changing the device underneath it and will have to have the placeholder device for safety purposes, but I believe that we can safely remove it until we encrypt the pool.

[mauelsha]

The reason for not keeping placeholder mapped devices with 1:1 linear mapping around in the stack is the overhead showing increasingly with growing I/O load/patterns which can be as much as 3-5% on fast devices. This kind of overhead has to be avoided in case it's not required. Device tree insertions to e.g. add caching can be applied if needed with device-mapper.

Relative to the crypt mapping concerns, the encrypted data has to be mapped identically when inserting e.g. the cache layer, i.e. at the table line argument of the "crypt" mapping, the start of the encrypted data has to be mapped with the same (or larger) previous size when a newly created mapped device (e.g. with cache, clone, integrity, raid, ... mapping) is inserted between the "crypt" mapped device and its previous backing (mapped) device thus replacing its <device_path> table line argument. I.e. deterministic block translation is strictly required in order to prevent data corruption.