Island pruning

[optionsome]

Summary

This is #2676 ported over to otp2. Additionally, the streetUtils methods have been moved to be under the PruneFloatingIslands class as suggested in the old pr.

Issue

#2675

Unit tests

To be added

Code style

Formatted the PruneFloatingIslands file to follow the code style convention

Documentation

It should be decided if this feature should be configurable as suggested in #2676 (comment) or not.

Changelog

To be added

[optionsome]

There are some things to discuss here:

• Should this feature be configurable as suggested in Island pruning improvements #2676 (comment) (and split into different modules)?
• How verbose should the logging be and what should be included in the HTML format report that can be enabled with dataImportReport=true in build-config.json?
• Should we include unit tests for this feature?

[optionsome]

We discussed this in today's meeting and decided that this new feature should be made configurable so it can be disabled as it increases memory consumption and build time. Also, I'll recheck if the current logging and issue storing works as intended and add some unit tests.

[vesameskanen]

I hand picked all relevant changes from both mfdz and hsldevcom into one singe commit, to get hopefully right content with decent amount of efforts. Some commit history was lost, sorry for that.

[vesameskanen]

A brief summary of changes:

• Bike, walk and car traverse mode connectivity is processed symmetrically
• Pruning happens now after stop-street linking, so island classification by stop existence actually works (has been broken before, also in OTP1)
• Now pruning might remove a small walk network around a linked stop. I tried rerunning stop-street linking after pruning but that seems to crash. Previously, those isolated stops at least sometimes linked to the continuous part of the graph, so this is a minor regression.
• Old code also included removal of transit links from a pruned island. That code got active after pruning was moved to a later stage, and seems to corrupt the graph (npe error at graph loading). I removed the crashing code - transit links stay alive around isolated stops, which seems harmless.

Current impementation works better than previous versions. Helsinki area data includes one stop whose walk network (2 edges) is removed by pruning. This is most likely an unusual data error which should be fixed.

There are 2 options to tackle the stop isolation problem:

• Drop island classification by stop existence (this has been the actual process in the past). The risk is that true small networks around stops get removed, for example a ferry stop on an island with only few paths/roads.
• Find out how to re-link isolated stops after pruning, without crashing the graph builder,

[t2gran]

Look good, only a few things since I last read through this.

[hannesj]

I'd like to see traverse modes (original, removed and resulting, as well as if it was a no through traffic island, but those can be added afterwards)