It really doesn't matter -- you can use either polling or event-based interrupts to simulate the other kind of input, so that decision is largely irrelevant in trying to mimic the characteristics of the input from older devices.
That said, you can find a lot of information about systems that old on emulation enthusiast websites (this one, for example, contains a wealth of information about the hardware and memory-mapped registers uses by the GameBoy -- it's similar to the document linked in some of the comments, if not exactly the same in a different format).
To achieve emulation of a particular game's input characteristics you'd want to do it at a higher level -- taking into account things like how the game handles two opposing directional bits set for joystick input, for example. Many older games don't handle this well, because it is generally not possible to physically interact with the device fast enough or in such a fashion as to cause some of the potential states to exist. This is how tool-assisted speed runners do some of their work.
You may want to scan through the TASVideos resource page for Pokemon and see if maybe there are any control-related oddities you should be emulating if you really want that level of similarity to the game.
But I don't think it matters very much to the end result if you choose polling or not, even though that is generally what was used on the GameBoy.