tomo: respect _rlnTomoVisibleFrames in particle reconstruction

[amineuron]

Summary

reconstruct_particle.cpp::processTomograms determines per-tilt visibility for each particle purely from Tomogram::determineVisiblity(traj, s/2.0) (geometric center-in-frame test). This silently ignores the explicit per-particle _rlnTomoVisibleFrames mask that subtomo writes into particles.star when --max_dose (or any frame filter) is set at extraction.

ml_optimiser (Refine3D / Class3D) already respects the mask — see src/exp_model.cpp "Add only the visible images for this particle". So in a --stack2d --max_dose pipeline, Extract and Refine3D use one frame set, but standalone relion_tomo_reconstruct_particle uses a different (geometric-only) frame set.

This patch makes reconstruction match the same authoritative pattern: when EMDL_TOMO_VISIBLE_FRAMES is present on the particle table, intersect geometric visibility with the explicit mask.

Rationale

When --max_dose filtered the 2D stacks down to e.g. 16.5 visible frames per particle on average, reconstruction was using ~39.2 frames per particle (all geometrically visible tilts including dose-excluded ones). The per-particle CTF + dose-weight math at reconstruction time therefore used a different frame set than:

1. The frames baked into the *_stack2d.mrcs data on disk
2. The frames Refine3D used for the alignment that we're reconstructing from

The symptom in 2D-stack mode is a "cloud-of-points" looking merged.mrc while data_*.mrc (the un-Wiener-divided numerator) still looks healthy. This is hard to attribute to anything in the user-facing inputs.

Compatibility

• Backwards-compatible: behaviour unchanged for inputs without _rlnTomoVisibleFrames. The has_explicit_visible_frames check only engages the new path when the column is present.
• No new dependencies; uses the existing ParticleSet::getVisibleFrames helper that exp_model.cpp already calls.
• Hard-errors with a clear message if the column exists but has an unexpected length for the tomogram.

Tests

• git diff --check
• Builds clean on top of ver5.1 (verified locally as part of the same fix in an unaffiliated fork; same one-file patch).
• On the project used for diagnosis (4 tomograms, ~6200 2D-stack particles, --max_dose 50): post-fix <isVisible> per particle drops from 39.2 to 16.5, matching the _rlnTomoVisibleFrames mask that Extract wrote.

🤖 Generated with Claude Code

[amineuron]

Worth noting: this fix is complementary to (and independent of) @daniel-ji's open #1305, which corrects the geometric-visibility radius from s/2.0 to binning * s / 2.0 in the same call. Both bugs live in the same line and both should be merged:

• fix visiblity determination for subtomo particle reconstruction #1305 fixes the box-size used by the geometric visibility test (post-bin vs pre-bin).
• This PR (tomo: respect _rlnTomoVisibleFrames in particle reconstruction #1344) adds the missing intersection with the explicit _rlnTomoVisibleFrames mask that subtomo writes when --max_dose is set.

I have both fixes stacked locally and they apply cleanly without conflict. Happy to rebase on top of #1305 if it's preferred to merge that one first.