Fixing bug in UsageGraph rendering.

The calculateLocalPaths() method of UsageGraph converts a set of paths
in (milliseconds, percent) coordinates into the actual pixel values that
will be used for drawing. For the most part this is a one to one
process, but not always: input points that are too closely spaced to
draw accurately are skipped. The last point in the path, however, is
never skipped, in order to ensure that the graph ends at the correct
location.

The previous implementation of this method had a bug: the y-coordinates
of points that were skipped would be stored indefinitely (in the local
variable pendingYLoc) and then added back at the very end of the path
(under the condition i == mPaths.size() - 1 && pendingYLoc !=
PATH_DELIM). Under the right conditions, this led to the strange uptick
at the end of the graph seen in the associated bug.

This CL fixes the problem and attempts to make the logic slightly
clearer. It also adds tests, one of which (_similarPointMiddle) fails
for the previous code.

In more detail, previously pendingXLoc was used to hold the last x
coordinate seen, while pendingYLoc was used to hold the last *skipped* y
coordinate, or PATH_DELIM otherwise. The difference between these was
somewhat subtle and hard to understand from a quick read of the code,
and there was a bug: pendingYLoc never got reset to PATH_DELIM even if
later points were added. In this CL I have removed the pendingLoc
variables in favor of a single lx/ly pair, which always holds the local
coordinates of the most recent point, and I added an explicit boolean
skippedLastPoint to track whether the point (lx, ly) has already been
added or was skipped.

Bug: 64065296
Test: make RunSettingsRoboTests
Change-Id: I45ccffea1280d851bfae5143c2e84d188e133731