Bring Galilean PIM in line with templated IMU preintegration

[dellaert]

Summary

• refactor the Galilean backend into GalileanPreintegration and define PreintegratedImuMeasurementsG as PreintegratedImuMeasurementsT<GalileanPreintegration>
• replace the separately implemented Galilean PIM with the common templated PIM API, including Boost, Python pickle, and MATLAB string serialization
• expose the Galilean PIM through the normal template/typedef wrapper machinery, using the alias-serialization support from Support serialization metadata on template typedefs wrap#208
• optimize Galilean integration, bias correction, residual, and Jacobian hot paths

Performance

The Galilean hot-path microbenchmarks improved approximately as follows:

• integration: 0.99 us to 0.56 us per sample
• factor residual: 158 ns to 115 ns
• factor residual with Jacobians: 1235 ns to 954 ns

Analytical outputs and Jacobians are unchanged.

Wrapper behavior

Python and MATLAB expose PreintegratedImuMeasurementsG through the same template/typedef machinery as the other preintegration backends. The generated wrapper includes the standard PIM methods and serialization without registering a duplicate native type.

Compatibility

• existing PreintegratedImuMeasurements, ImuFactor, and ImuFactor2 construction, serialization, naming, and APIs remain supported
• the NEES notebook was rerun and produced unchanged results, so its output is not included

Testing

• testGalileanImuFactor.run
• testSerializationNavigation.run
• targeted Python IMU, Galilean, gravity, and serialization tests in the py312 environment
• Python extension and stub targets built
• gtsam_matlab_wrapper generated and compiled
• wrap test suite: 134 passed
• git diff --check

Closes #2756

[dellaert]

@rohan-bansal will merge but if you see any issues just comment and ping me on slack.