Make view graph calibration a stand-alone estimator module on colmap::Database and add CLI.

[B1ueber2y]

This PR includes:

• Make view graph calibration work on database instead of reconstruction and ViewGraph. Add CLI.
• Move view graph calibration into colmap/estimators. Add nominal tests that can pass with the current cost functor.
• Move view graph calibration out of the global mapper into GlobalPipeline, to be further moved out as a stand-alone module inside automatic reconstructor in the future.
• With the simplified global mapper, we can remove the dependency of global mapper in the tests of rotation averaging and gravity refinement and move them to glomap/estimators.

[ahojnnes]

Extract 1-e3 as a constexpr?

[ahojnnes]

Surprised we are not setting an upper bound.

[ahojnnes]

Also wondering whether it makes sense to bound the parameter using options.min_focal_length_ratio/options.max_focal_length_ratio?

[B1ueber2y]

Lets for now extract 1e-3 as a constexpr and keep it as it is, to keep the logic equivalent, as bounding the parameter directly may end up not reverting those diverged focals (not sure if there are any). Maybe this can be resolved by checking whether the focal lengths are exactly at the boundary after optimization, but this needs some studies and goes beyond the scope of this PR.