fix(gui): omit --frames for entire-video inference (#2808)

[tom21100227]

Problem

Running inference from the GUI with the frame range set to "Entire video" builds a command containing --frames 0,-N (e.g. --frames 0,-10245). This was reported in discussion #2807, where it crashes with IndexError: Failed to read frame index N.

Two things are wrong with the 0,-N encoding:

1. It's fragile. MainWindow._get_frames_for_prediction encodes "Entire video" as the half-open range [0, len(video)) → [0, -len(video)], and VideoItemForInference.cli_args serializes that to --frames 0,-N. That string only decodes back into a range because sleap-nn's frame_list happens to do min_max[0].rstrip(","). It reads like the two-frame list [0, -N] (exactly how the reporter interpreted it).
2. It pins the endpoint to the GUI's frame count. If that count is one greater than the number of frames the inference reader actually decodes — the 1-indexed/0-indexed discrepancy in GUI frames are 1-indexed, while cli commands are 0-indexed #2790, or a video whose metadata over-reports its length — inference iterates to a nonexistent frame and raises IndexError: Failed to read frame index N.

Fix

Add VideoItemForInference.is_entire_video and, when it's true, omit --frames so the inference CLI predicts on all frames using its own frame count (the documented default when --frames is not passed). This drops the fragile encoding and decouples the run from the GUI's count.

Selection	Before	After
Entire video	--frames 0,-N	(omitted → all frames)
Clip [a, b)	--frames a,-(b-1)	unchanged
Explicit frames	--frames 0,1,2,…	unchanged
Current frame	--frames k	unchanged

Testing

• tests/gui/learning/test_cli_construction.py::TestVideoItemForInferenceCLI — added 3 tests (entire-video omits --frames; a shorter clip still emits --frames; explicit lists are never treated as entire-video). All 9 tests in the class pass.
• Existing frame-range / CLI-construction tests unchanged and green.

One unrelated test (TestWritePipelineFiles::test_train_script_does_not_contain_zmq) fails locally with ModuleNotFoundError: No module named 'sleap_nn' — pre-existing in this environment (confirmed on a clean branch), not touched by this change.

Notes

If the crash still reproduces for a specific video after this change, the video's own metadata over-reports its decodable frame count (a corrupted / variable-frame-rate file). That is a separate reader-robustness concern in sleap-nn / sleap-io, and the plan is to confirm with the reporter and request the video if needed.

Fixes #2808. Reported in #2807.

[github-actions]

Docs Preview

Preview URL	https://docs.sleap.ai/pr/2809/
Commit	9a5f1d00e6b93b85a7afcacd775911239b3b535f

API changes detected - View API Reference

This preview will be removed when the PR is closed.

[codecov]

Codecov Report

❌ Patch coverage is 83.33333% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 70.70%. Comparing base (2649d80) to head (9a5f1d0).

Files with missing lines	Patch %	Lines
sleap/gui/learning/runners.py	83.33%	3 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #2809   +/-   ##
========================================
  Coverage    70.69%   70.70%           
========================================
  Files          103      103           
  Lines        23359    23374   +15     
========================================
+ Hits         16514    16526   +12     
- Misses        6845     6848    +3     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.