Fix the troposphere phase screen datacube across the dateline

[xhuang-jpl]

Problem: The AOI bounding box used plain min/max over radar-grid longitudes. For scenes crossing ±180°, wrapped longitudes (e.g. +179°/−179°) inflate the AOI to ~358° wide, so RAiDER loads a near-global weather model and the delay cube is wrong for dateline-crossing acquisitions.

Fix: Wrap raider_tropo_delay in compute_tropo_delay_dateline, which detects the crossing and only then splits the output grid into east/west sub-AOIs, runs each against its own weather-model crop, and merges the cubes back along x (reassembled in original column order). Non-crossing scenes take a single call with unchanged behavior.

Runtime: It takes about 64 seconds on the c5.4xlarge instance, and the plot looks like

[bhawkins]

Thanks for the patch. The approach makes sense, though it has some limitations.

[oberonia78]

I'm not sure why tropo_delay_fn is introduced. Previously, it seems that you used the radar_tropo_delay at default, and I don't see the usage of tropo_delay_fn.

[xhuang-jpl]

it is a parameter of the compute_tropo_delay function and its default is radar_tropo_delay, I can delete it.

[bhawkins]

LGTM, thanks. Hopefully we can address the remaining gap in a future PR.

[hfattahi]

We reviewed in a PR review and these changes look good to me.
Please open an issue for the remaining artifact at the dateline boundary.