Others have already got it right, but I'd like to present a more puzzle-like solution:
Going through the numbers 2-47 once produces some deck state $Y$ (some cards face-up, others face-down) from a starting deck state $X$.
Because we're given a clearly defined sequence of reversible operations, it is possible to uniquely determine $X$ given $Y$, and, of course, vice versa.
Therefore, every deck state has a unique successor, and a unique predecessor.
This allows us to imagine that every possible deck state is a snap shackle, so that the openable part (the shackle) ties the state to its successor state's permanently closed part (the ring):

Now, we have quite a bunch ($2^{52}$) of these shackles, and we are going to completely ignore what the actual transformation does. Instead we'll just connect the snap shackles to each other so that
- no ring is left without a shackle (every state has a predecessor)
- no shackle is left without a ring (every state has a successor)
- two shackles cannot connect to the same ring (the predecessor is unique)
- no shackle can connect to more than one ring (the successor is unique)
Since rules 3 and 4 prevent any kind of bifurcation, the only way to fulfil requirements 1 and 2 is to build one or more closed loops. This means that every single shackle has to be a part of a simple chain loop, regardless of what the actual transformation rules are.
Or coming back out of the analogy: If you repeat any transformation (in this case, going through numbers 2-47) many enough times, you are guaranteed to eventually end up in the original deck state.