[nodes] new SfMTriangulation node

[mh0g]

Description

Adds a main that is essentially the triangulation step from the sequential SfM.
AliceVision PR: alicevision/AliceVision#1314

Features list

This allows to run the triangulation without going into the sfm step.

[mugulmd]

When testing this node it kept failing with the following error:
ERROR: unrecognised option '--maxMemory=9223372036854771712'

[mugulmd]

Tested on 3 datasets by using the output of the StructureFromMotion node as input of the Triangulate node.
Results:

• the generated landmarks are similar to the ones generated by StructureFromMotion, with a few differences (which is expected since there is no bundle adjustment)
• when increasing the minNumberOfObservationsForTriangulation the number of landmarks decreases (which is expected).

I also ran the following sanity check: when using an SfMData without poses as input, no landmark is generated.

[mh0g]

Tested on 3 datasets by using the output of the StructureFromMotion node as input of the Triangulate node. Results:

* the generated landmarks are similar to the ones generated by `StructureFromMotion`, with a few differences (which is expected since there is no bundle adjustment)

* when increasing the `minNumberOfObservationsForTriangulation` the number of landmarks decreases (which is expected).

I also ran the following sanity check: when using an SfMData without poses as input, no landmark is generated.

Tested on windows, seems to work as intended.
Tested on a benchmark dataset with ground truth poses and features/matches from meshroom. Small differences vs running with poses from SfM (as expected). Can also confirm that the landmarks are recomputed when using the ouptut of the Sfm node.

[fabiencastan]

It would be good to also add a node for https://github.com/alicevision/AliceVision/blob/develop/src/software/pipeline/main_computeStructureFromKnownPoses.cpp
Or fuse them in a single node.

[fabiencastan]

After checking, no need to integrate computeStructureFromKnownPoses, already possible with FeatureMatching+Triangulation.