[WIP] New photometric VS features + improvements to vpFeatureLuminance

[SamFlt]

This PR introduces new visual servoing features, taken from the papers of Eric Marchand and readapted from his code.

These features are PCA and DCT compressed representations of the luminance information.

In terms of implementation, new classes are introduced:

• vpFeatureLuminanceMapping a visual servoing feature, which combines a vpFeatureLuminance object with a vpLuminanceMapping subobject. This is the feature that is given to vpServo
• vpLuminanceMapping, a common interface for mappings that operate on luminance information. a map method computes a vector representation from the image. The inverse method, converts this vector back to an image, with some information loss. Finally, the interaction method computes the Jacobian, given the image, the compressed representation and the photometric Jacobian (computed beforehand by vpFeatureLuminance when used with the vpFeatureLuminanceMapping)
• The vpLuminanceMapping interface has two implementations: vpLuminancePCA and vpLuminanceDCT. they can be used interchangeably in vpFeatureLuminanceMapping

Before this PR is complete, several things are lacking:

• Encapsulate a Levenberg-Marquardt control law in a custom class.
• Add weighting mask when computing the error in vpFeatureLuminance
• Add an option to have a depth per pixel instead of considering that all 3D points are at an equal distance from the camera.