IR1 bug fixes

[wbenoit26]

@deepchatterjeeligo Fixes a few bugs I encountered while running over the LLPIC. None of these are strictly necessary, but are nice to have. All of the commits are self-contained, so we can cherry-pick a subset if desired.

• The binning in the latency histogram caused the fastest and slowest events to be excluded from the plot. I've added a bit of padding to the bin range.
• When the CI builds the docker image, the LD_LIBRARY_PATH set in apptainer.def doesn't get transferred over, which can break imports. The search happens to make it through, but the monitor fails without the environment variable set.
• I noticed that our latency measurement includes the GraceDB response time. I've added another timestamp to record the time it takes to get a response back, so the histogram just shows latency coming from our end.
• When the node rebooted last time, the process got shut down in a way such that the final segment got written to segments.txt but the current.json heartbeat file didn't get removed. When the search restarted, the flush of the old current.json caused the final segment to get duplicated. I've added a check before a segment gets appended for whether the current segment has the same state and start time as the previous one, and also a function to check segments.txt for overlapping segments and write a warning to the summary page if any exist.

[deepchatterjeeligo]

Had a look. All changes make sense to me. Happy to give this a shot