I agree, but (as you observed) it seems to have many faithful fans.
In my personal symbol library all of the 3-terminal regulators have the output pin defined as “Power output”, and both the input pin and the ground pin are defined as “Power input”. (The Enable pin, if present, is an “Input”.) These assignments are correct as I understand the ERC operation. I don’t know how these pins are defined in the “standard” KiCAD libraries, or whether they are defined the same way for ver4 and ver5 libraries.
As I understand the “power flag” symbol, it is used to mark nodes which (under normal operation) are involved with supplying power to the circuit. For example, the terminals of a battery are inherently capable of supplying power, so they are defined in the battery symbol as “Power output” and the schematic nodes they connect to do NOT require a “Power flag” symbol. On the other hand, a connector - such as a two-terminal pin header - is not inherently a power source. The pins on its schematic symbol should normally be “Passive” or “Unspecified”. When you put this symbol on a schematic and use it as a power-input connector, it is necessary to place a “Power flag” on the nodes it connects to. This tells the ERC that these nodes are potentially sources of power to the circuit.
As @1.21Gigawatts pointed out, things become complicated when you add components in series with a power rail - such as switches, fuses, OR’ing diodes, common-mode chokes, etc. Then you have to think through the “power flag” problem, and thinking always makes my brain hurt.
The practice of defining a pin’s “Electrical type”, and applying “Power flags” where needed, comes from the desire to perform a rudimentary ERC. I’m not convinced it’s worth the effort; in my work, it has produced many more false alarms than the number of true problems it has alerted me to. However, I continue to use the tool in the expectation that some day it will demonstrate its value.