Fix TimeFilter for very small time intervals with tracklength estimator

[paulromano]

This PR fixes #2156. The issue is as follows: if you have a source with a large time specified, e.g., 1.0e8 seconds, the time interval over a single track might be so small that the starting and ending time for that track are effectively the same. This is just a result of floating-point arithmetic because it's possible to have $t + Δt = t$ in if $Δt$ is small relative to $t$. Right now, such tracks will not get marked as match for a time filter even if the starting time is within a specified time bin. This is now fixed in TimeFilter::get_all_bins and for good measure, I've added a test that checks this particular condition.

Thanks to @tibetscarlett for identifying the issue and giving a proposed solution!