Because the previous implementation of determining the time the user
falls asleep in a given time range would take the 24 hours in advance
into account, graphs displaying sleep data would erroneously indicate
that the user has been asleep since the start of the timeframe if
the user was asleep during the rollover of the time frame 24 hours
before.
This commit change the algorithm to only fetch the last sleep stage
sample and sleep range sample from the database that occurred before
the given time range. This saves having to process 24 hours worth of
samples before the time range in both cases, and prevents taking into
account irrelevant sleep ranges.
