changed barcsq to a vector to allow each factor to have a different inlier threshold

[lucacarlone]

I realized that having a default inlier threshold equal to 1 does not make sense for the gtsam user since:

1. "1" is not a reasonable error threshold (for Gaussian noise, we should use the chi2inv to establish a reasonable maximum error) and
2. the factors can have different dimensions, hence it is desirable to be able to set a different threshold for each factor.

The PR involves:

• making barcsq into a vector
• adding/adjusting the unit tests accordingly
• initializing barcsq using chi2inv

[lucacarlone]

@dellaert @jingnanshi @pantonante : I completed a small enhancement for GNC. please take a look and approve if you like it.

@dellaert : please take a look at the 2 lines commented at line 658 of testGncOptimizer. It seems that "dim()" segfaults when we use the default noise model in the factor. This might be a bug in gtsam.

[dellaert]

Nice! Sorry about delay, teaching large class ;-)

[dellaert]

PR is fine, but changing to "request changes" as I just noticed CI fails on Linux. Probably something small but needs to be fixed before we can merge.

[dellaert]

@dellaert : please take a look at the 2 lines commented at line 658 of testGncOptimizer. It seems that "dim()" segfaults when we use the default noise model in the factor. This might be a bug in gtsam.

Is that why CI fails? Let me check.

[lucacarlone]

@dellaert: let me know if you get some insight into what's failing in CI. I took another pass and cleaned up the comments. @jingnanshi @pantonante : feel free to approve if this looks good.

[jingnanshi]

We might want to remove this line.

[lucacarlone]

good catch - done!

[jingnanshi]

Everything looks good! I left one comment regarding a commented out include file.

[dellaert]

Still some failures on Ubuntu, though?

[lucacarlone]

@dellaert : I only changed comments so far, I hoped you could look into the dim() issue. Let me know if that's not feasible.

[dellaert]

@dellaert : I only changed comments so far, I hoped you could look into the dim() issue. Let me know if that's not feasible.

Dim issue is separate, I think. CI failure is unrelated

[lucacarlone]

@dellaert : @jingnanshi fixed the issue! at this point, this branch should be good to go