memory: tegra20-emc: Poll EMC-CaR handshake instead of waiting for interrupt

The memory clock-rate change could be running on a non-boot CPU, while the
boot CPU handles the EMC interrupt. This introduces an unnecessary latency
since boot CPU should handle the interrupt and then notify the sibling CPU
about clock-rate change completion. In some rare cases boot CPU could be
in uninterruptible state for a significant time (like in a case of KASAN +
NFS root), it could get to the point that completion timeouts before boot
CPU gets a chance to handle interrupt. The solution is to get rid of the
completion and replace it with interrupt-status polling.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>