Reset the base when pull fails.

Without resetting the base, we will compute the wrong diff when we have
a condition change.

Move the responsability of handling failures to the PullReceivers. This
is more consistent with #onConditionChanged which does handle the
failure cases.

We also always compute nextPullTimeNs in order to only call onDataPulled
on bucket boundaries. The current code does not update nextPullTimeNs
which means a new alarm might trigger a pull and onDataPulled in the
middle of the bucket. The behavior in this case is undefined.

Bug: 123181864
Test: atest statsd_test
Change-Id: I0910b7db26a0de764436c46c8d7d11cafa52dcd9