Instead of just printing an error when mmc_of_parse() fails to request
CD/WP GPIO pins, let's propagate all errors, except for -ENOENT.
Earlier only -EPROBE_DEFER was handled correctly.

As a side effect of this change and by relying on host drivers to
handle the errors during ->probe(), we don't need to free any data in
the error path.

This also means we are actually fixing a bug, since we remove the call
to mmc_gpio_free_cd() which wasn't the correct function to invoke to
handle cleanup. Instead that should have been mmc_gpiod_free_cd().
Signed-off-by: NUlf Hansson <ulf.hansson@linaro.org>