Also confirming: Horses are still useless in 15w46a.
Confirming this bug is affecting 15w45a
Confirming for 15w45a, too. Also, unloaded/un-rendered chunks remain unloaded/un-rendered even after debug chunk reloading (F3 + a). Affected chunks do load and render almost immediately upon dismounting from horse.
This bug can be worked around without starting a new world using NBTExplorer or other NBT editing software. Open your world's level.dat, and go to Data, DimensionData, 1, DragonFight, and set DragonKilled and PreviouslyKilled to 0.
NOTE: This work-around is probably NOT officially supported or recommended by Mojang. If you decide to do this, you do so at your own risk.
I'm experiencing this bug too, but in a village I created from scratch by curing zombie villagers. Some of these reports list only the top half of the doors going missing, but I'm getting the bottom half disappearing.
This bug also manifests in another way... On my SMP server, the bug sometimes triggers while I'm in the village, but the server doesn't seem to notify my client that the doors were removed; So I can still see both halves of the doors, and I can even open and close them. However, if I try to go through a glitched door while my client thinks it's open, I'll start rubber-banding and the log fills up with "Player has moved wrongly" messages. If I then break the glitched door, I will get a door item back, which I then replace the door, and it's fixed for a while.
As others have mentioned, this bug appears to occur every time the affected chunk is loaded. As for when it occurs while the chunk is already loaded, it seems like it may have something to do with when a villager opens, closes, or goes through a door, but not every time; So I may be wrong about this relation.
I am also experiencing this issue in survival mode.
This bug also duplicates the glitched rails when broken and picked up.
Rails do not update their power status upon placing with any Redstone source, not just Redstone Blocks.
When this bug occurs, the rails placement sound does not play.
This bug does NOT occur when placing rails of the same type adjacent to each other; only occurs when placing rails of different types adjacent to each other.
OS: Manjaro Linux, x86_64
Java: Oracle JDK 1.8.0u51, x86_64
Still an issue in 15w46a