Preserve IMOD final tilt angles in YTilt

[amineuron]

Summary

• keep IMOD final .tlt angles in _rlnTomoYTilt instead of subtracting AngleOffset
• make the C++ align_tiltseries runner and Python IMOD helper consistent
• document that .tlt is IMOD's final OutputTiltFile used for reconstruction geometry

Rationale

When IMOD tiltalign estimates AngleOffset, the final .tlt file already includes that offset and IMOD tilt reconstructs from those corrected angles. Subtracting AngleOffset while writing _rlnTomoYTilt reverts the STAR metadata to nominal stage tilts, making RELION projection matrices inconsistent with the IMOD reconstruction geometry. This matches the symptom reported in #1141.

CTF / downstream extraction

This PR intentionally does not rewrite existing _rlnCtfScalefactor values during alignment. RELION native tomogram reconstruction already initializes _rlnCtfScalefactor = cos(_rlnTomoYTilt) when the column is absent, so preserving the final IMOD angles keeps the default CTF scale consistent with the projection geometry.

If _rlnCtfScalefactor is already present, it may have been written by RELION CTF refinement as a signal/ice-thickness scale. Overwriting it at alignment-import time would be unsafe.

Tests

• git diff --check
• python -m py_compile src/tomography_python_programs/align_tilt_series/imod/_utils.py

[amineuron]

Closing this PR.

After deeper investigation in my fork, I could see you might had designer choice and might need to be sure this convention change would be safe, couldn't find why Relion choose different convention

1. Whether _rlnTomoYTilt stores IMOD's pitch-corrected angle or the nominal angle is empirically equivalent for self-consistent project chains — every YTilt consumer in the built tomo code (Tomogram::getCtf, reconstruct_tomogram.cpp, subtomo.cpp, reconstruct_particle.cpp) uses whatever YTilt is in the STAR.

2. For our own fork the IMOD-consistent convention is convenient, but I don't think it's the right thing to push upstream, it would change default behavior for every existing user without solving a reproducible bug.

Will submit a focused PR for the _rlnTomoVisibleFrames issue (reconstruct_particle.cpp uses geometric visibility only, while exp_model.cpp correctly intersects with the explicit mask, so Reconstruct and Refine3D end up using different frame sets when --max_dose was set at Extract).

Thanks for the review patience.