Fix Varian loader direction and XIM parsing

[DebadityaHait]

Summary

• Correct bowtie calibration-file selection for clockwise Varian TrueBeam 2.7 scans.
• Decode uncompressed XIM pixel buffers into shaped signed-integer arrays.

Real-data validation

Using TIGRE’s documented Varian CBCT resolution-phantom dataset:
https://zenodo.org/records/8045749

• Base: StartAngle=180.0, StopAngle=-20.0 selects CC and loads 10 FilterBowtie_CC_* files.
• This PR: the same scan selects CW and loads 10 FilterBowtie_CW_* files.
• The sample calibration XIM files use compression=1.

[AnderBiguri]

have you tested this with Varian data? can you show me? or is this an AI random PR?

[DebadityaHait]

Yes—I used AI assistance to prepare this PR, but it is a targeted parity fix rather than a speculative change.

I tested it with TIGRE’s documented Varian CBCT resolution-phantom dataset: https://zenodo.org/records/8045749

Its TrueBeam 2.7 Scan.xml has StartAngle=180.0 and StopAngle=-20.0. On that dataset, the parent version classifies the scan as CC; this PR classifies it as CW and load_blank_projections() successfully loads all 10 FilterBowtie_CW_* calibration images.

I also added regression coverage for both angle signs and for the FilterBowtie_CW*.xim selection path.

[AnderBiguri]

Thanks! can you put both of your PRs here, remove the test files, and show me the before/after of the test, if anything changes?

[DebadityaHait]

Consolidated #764 into this PR and removed the PR-specific test files as requested.

Real-data before/after using TIGRE’s documented Varian CBCT resolution-phantom dataset:

Version	Direction	Loaded blank calibrations
Base	CC	10 FilterBowtie_CC_* files
This PR	CW	10 FilterBowtie_CW_* files

The sample calibration XIM files use compression=1, so the uncompressed-XIM branch does not change output for this dataset.

[AnderBiguri]

can you show the reconstruction differences before/after? To validate the code does indeed change the recon.

[DebadityaHait]

I ran matched FDK reconstructions on TIGRE’s documented Varian CBCT resolution-phantom dataset.

Both runs used 32 evenly spaced real projections, a fixed 128³ reconstruction geometry, and disabled detector/scatter corrections.

Version	Blank calibrations	Reconstruction
Base	10 FilterBowtie_CC_* files	Base volume
This PR	10 FilterBowtie_CW_* files	Corrected volume

The reconstructed volumes differ:

• Mean absolute difference: 0.0004847554
• Maximum absolute difference: 0.0045613390
• Relative L2 difference: 4.586146%

The embedded central-slice comparison shows the base volume, corrected volume, and their absolute difference.