So you state here:

Explicit conversions to and from address are allowed for uint160, integer literals, bytes20 and contract types.

Later below, in the "Warning" section you state:

You can use address(uint160(bytes20(b)))...

I also have seen the address(uint160(bytes20(...))) patterns many times in codebases. However, the bytes20 to uint160 cast is unnecessary as you even state in your docs that you can explicitly convert from bytes20 to an address. You can simply test this:

➜ bytes32 a = keccak256("hello")
➜ address(uint160(bytes20(a)))
Type: address
└ Data: 0x1C8AFf950685C2Ed4BC3174F3472287b56D9517b
➜ address(bytes20(a))
Type: address
└ Data: 0x1C8AFf950685C2Ed4BC3174F3472287b56D9517b
➜ payable(address(bytes20(a)))
Type: address
└ Data: 0x1C8AFf950685C2Ed4BC3174F3472287b56D9517b

Patched here: #16132.