ci: align E2E cluster node sizing across cloud providers (#10687)

GKE nightly E2E fails intermittently when a replica pod cannot be
scheduled: the zonal PD is pinned to a node that the kubelet has tainted
under disk pressure caused by the container image cache filling the 20
GiB boot disk. AKS occasionally picks Standard_D4plds_v5 (arm64) when
--node-vm-size is unset, mixing architectures across the matrix and
reducing test consistency.

Pin all three managed providers to x86_64 4-vCPU SKUs and raise GKE's
boot disk to match EKS: e2-standard-4 with 80 GiB on GKE,
Standard_D4s_v5 on AKS, m6a.xlarge on EKS.

Closes #8331

Signed-off-by: Marco Nenciarini <marco.nenciarini@enterprisedb.com>