New pages from buffer pool, that belong to encrypted space can be mistaken for unencrypted ones by encryption threads

When page is encrypted it gets key version assigned before it is written to disk. However when page is created in buffer, even though it belongs to encrypted space it will not get key version assigned, only after syncing to disk. Encryption threads can take this page for unencrypted one and report there are unencrypted pages in the tablespace and thus mark whole tablespace as unencrypted. The best way would be to assign key version for each new block created in a space, but I would need to cover all the paths from space to buf - to propagate the key version, and that would be a big diff. Thus it possibly be better if encryption threads knew if that page was created in a buffer and assign the proper key version to that page.