(1) Base64 of SHA256 seems overkill. 256 bits of hash is a lot. The risk here is that a MITM intercepts the payment request, which will be typically requested just seconds after the QR code is vended. 80 bits of entropy would still be a lot and take a long time to brute force, whilst keeping QR codes more compact, which impacts scannability.

(2) This should not be necessary in the common HTTPS context. The QR code itself is going to be fetched from some service, over HTTPS. I see no reasonable attacker that can MITM the request for the BIP70 message but not the request to get the QR code. Adding a hash makes QR codes more bloated and harder to scan, all on the assumption that HTTPS is broken in some odd way that we haven't actually ever seen in practice.

(3) This can be useful in the Bluetooth context, but then again, we could also do things a different way by signing with the key in the first part of the URI, thus avoiding the need for a hash.

I know I've been around the loop on this one with Andreas many times. But this BIP doesn't fix any actually existing problem in the previous spec. It exists because Andreas thinks SSL is useless. If SSL is useless we all have much bigger problems.