Use millisecond precision in A* traversal

[tkalvas]

Summary

The existing State handling code in searches updates the time (integer, seconds) by calculating the time used by each step, rounding it up to the nearest integer, and then updating the total. In well-mapped indoor areas, the steps are small, and there are many of them. In our example problem spot, there are 42 edges in 284 metres. This ends up making the total 35 seconds too long. Our transfers try to be well designed, and we have a 7 minute window for that transfer. It takes 287 seconds to walk that distance, and a 90 second transferSlack. We cannot afford inaccuracies of the order of half a minute.

This PR changes the unit of calculation to milliseconds in the actual traversal, and still uses whole numbers. The granularity of time elsewhere in OTP still mostly stays as one second, to avoid a massive change.

The symmetry of backward and forward changes is preserved. When the same trip is searched backward and forward, the duration is exactly the same both in milliseconds and when rounded to seconds. When a forward search is made, the rounded end time is taken as a start time of a reverse search, the rounded end time of the reverse search is the same as the start time of the forward search, given that that had a millisecond part of zero.

Unit tests

In a large number of tests trip durations have become slightly shorter. This accounts for the large amount of changes to tests.

Changelog

This will be a significant change, because the durations have been incorrect for a long time, and people will notice.

Bumping the serialization version id

Unnecessary.

[codecov]

Codecov Report

Attention: Patch coverage is 85.41667% with 7 lines in your changes missing coverage. Please review.

Project coverage is 70.35%. Comparing base (7f5c12c) to head (1cdaca7).
Report is 218 commits behind head on dev-2.x.

Files with missing lines	Patch %	Lines
...entripplanner/street/search/state/StateEditor.java	55.55%	1 Missing and 3 partials ⚠️
...tripplanner/street/model/edge/ElevatorHopEdge.java	50.00%	1 Missing and 1 partial ⚠️
...org/opentripplanner/street/search/state/State.java	92.85%	1 Missing ⚠️

Additional details and impacted files

@@              Coverage Diff              @@
##             dev-2.x    #6436      +/-   ##
=============================================
+ Coverage      69.84%   70.35%   +0.51%     
- Complexity     18124    18176      +52     
=============================================
  Files           2069     2062       -7     
  Lines          77268    76935     -333     
  Branches        7855     7764      -91     
=============================================
+ Hits           53965    54128     +163     
+ Misses         20546    20051     -495     
+ Partials        2757     2756       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[leonardehrenfried]

Use the seconds method here.

[leonardehrenfried]

This is not fixed. Please use the seconds method

[leonardehrenfried]

I think this PR will be a bit easier to review if you use the seconds method wherever possible and only millis where required.

[leonardehrenfried]

There is one more call to fix.

[optionsome]

Btw does this need to be long?