Buffer implied pointsets, and allow sub-pointsets to be uncontained b…

[mattwigway]

This does two things:

1. Buffers the implied WebMercatorGridPointset so that it expands beyond the street network extent (Buffer transport network when creating linked grid pointset #224). I used the max offstreet walk distance rather than the distance table size, because if points are further than that from the extent of the street network they are perforce unusable.
2. Allows sub linked pointsets to not completely overlap the superpointset. Currently generating Browsochrones data near the edge of the graph fails because the sub point set created around the origin to report non-transit times have points outside the implied pointset. This way those points outside the graph pointset are simply left unlinked. The reason this is a problem now and not in the past is that we recently made a change to simply subset the implied linked pointset, rather than generating a new linked pointset, when computing the non-transit times. Thus, we won't see walk paths to areas outside the extent of the implied pointset anymore, but this seems like a small loss - not linking a new, small grid pointset was a performance win.

[abyrd]

Looks good to me, added a couple of comments. The main thing I notice is that Github really mangles commit subject lines longer than about 55 characters, and while the commit messages can be expanded (breaking awkwardly in the middle of words) the title of the pull request has beed derived from one of the commits and cannot be expanded:

[abyrd]

This could also be done with

        Envelope envelope = new Envelope(transportNetwork.streetLayer.envelope);
        envelope.expandBy(lonBuffer, latBuffer);

[abyrd]

Maybe we should be using a symbolic constant like UNLINKED for this, but I guess we have hard-coded -1s all over.

[abyrd]

@mattwigway I'm going through these older pull requests, hoping to merge some of them. Is this one still applicable? Will there be any side effects to merging it (interference with other changes in the front end for example)?

[abyrd]

Revisiting this in October 2020, after combining r5 with analysis-backend. These changes may still be useful but my previous comment still stands - it will take some time to determine whether this is applicable or will necessitate changes elsewhere in the system.

[abyrd]

Accidentally closed when cleaning up branches. I would like to revisit these old pull requests and determine whether they're still relevant.

[ansoncfit]

This may be superseded by #684 and #688