Baseline-dependent rebinner

[ljgray]

Summary

Implements a rebinning task similar to #266. This updated version uses the actual per-baseline weights, rather than a mean, and avoids creating a separate rebin_weight dataset entirely. This should provide marginally better rebinning behaviour in the presence of bad baselines, and does not require complicated tracking of multiple datasets throughout various pipelines.

It also implements a baseline-dependent effective_ra dataset and a simplified effective_ra centring correction.

From testing, this seems to show a light reduction in noise across all delays compared to the mean-weighted rebinning.

Notes

• The rebinning has been tested through the chime daily pipeline
• Gradient correction has been tested
• Stacking of daily data has not been tested, but is assumed to be ok given no issues with daily products
• The addition of a stack axis in the effective_ra dataset means that containers produced by the old rebinner cannot be loaded with this PR

This PR will replace #266 and would close chime-experiment/ch_pipeline#242 as it is no longer needed.

Depends on radiocosmology/caput#239 and #231 for updates to container copying behaviour. I've decided to implement this in a future PR

[ssiegelx]

This looks great and simplifies things considerably compared to the baseline-independent weighting. I just have a few comments.

[ssiegelx]

Looks good!