hyperdrive panics when reading a 256T uvfits file

[gsleap]

Not a super urgent issue as we aren't close to 256T... yet!

Works perfectly for a 256T CASA Measurement set.

Example obsid: 1352408368
It is a 128T obs, but we have created a 256T metafits file and inflated the gpuxbox files to make them contain 256T worth of visibilities and weights. Data is available on mwax26- /visdata/calvin_test/1352408368_256t

Running hyperdrive with this cmd line:
/home/mwa/mwa_hyperdrive/target/release/hyperdrive di-calibrate --no-progress-bars --data 1352408368.uvfits 1352408368.metafits --num-sources 99 --source-list /home/mwa/srclists/GGSM_updated.fits --source-list-type fits --outputs 1352408368_solutions.fits 1352408368_solutions.bin

Produces this output:

...
[2024-10-04T06:12:36Z INFO ]
[2024-10-04T06:12:36Z INFO ] DI calibration set up
[2024-10-04T06:12:36Z INFO ] ├ 1 calibration timeblocks, 768 calibration chanblocks
[2024-10-04T06:12:36Z INFO ] │ 28 timesteps per timeblock
[2024-10-04T06:12:36Z INFO ] ├ Calibrating with 32184 of 32640 baselines
[2024-10-04T06:12:36Z INFO ] │ Minimum UVW cutoff: 50λ (126.607m)
[2024-10-04T06:12:36Z INFO ] │ Maximum UVW cutoff: ∞
[2024-10-04T06:12:36Z INFO ] │ (Used obs. centroid frequency 118.395 MHz to convert lambdas to metres)
[2024-10-04T06:12:36Z INFO ] ├ Chanblocks will stop iterating
[2024-10-04T06:12:36Z INFO ] │ - when the iteration difference is less than 1e-8 (stop threshold)
[2024-10-04T06:12:36Z INFO ] │ - or after 50 iterations.
[2024-10-04T06:12:36Z INFO ] │ Chanblocks with an iteration diff. less than 1e-4 are considered converged (min. threshold)
[2024-10-04T06:12:36Z INFO ] └ Writing calibration solutions to: 1352408368_d2d_solutions.fits, 1352408368_d2d_solutions.bin
[2024-10-04T06:12:36Z INFO ]
[2024-10-04T06:12:36Z WARN ] Warnings
[2024-10-04T06:12:36Z WARN ] ├ Will overwrite the existing file '1352408368_d2d_solutions.fits'
[2024-10-04T06:12:36Z WARN ] └ Will overwrite the existing file '1352408368_d2d_solutions.bin'
[2024-10-04T06:12:36Z WARN ]
[2024-10-04T06:12:49Z INFO ] Reading input data and sky modelling
thread 'read' panicked at src/io/read/uvfits/mod.rs:775:46:
no entry found for key
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
./hyperdrive_d2d.sh: line 1: 860262 Aborted                 (core dumped)

The line in question is:

let (ant1, ant2) = (self.tile_map[&uvfits_ant1], self.tile_map[&uvfits_ant2]);

Using Hyperdrive v0.4.1.
uvfits file was created with Birli 0.15.1 using:
/home/mwa/Birli/target/release/birli --metafits 1352408368.metafits --no-draw-progress --uvfits-out=birli_d2d/1352408368.uvfits --flag-edge-width=80 --max-memory=450 --avg-freq-res=40 --avg-time-res=2 *.fits

The uvfits file and the metafits seem to have the same list of 256 tiles. So not sure what else could be wrong without further investigation.