Update arrakis dataloader

[wbenoit26]

@deepchatterjeeligo @bhgupta20 I haven't been able to test yet because our robot credential needs the arrakis.read scope added to it, but would be good to get a couple sets of eyes on this.

Changes:

1. Renames NGDD to Arrakis
2. Makes the Arrakis dataloading more robust, with fewer magic numbers
3. Updates the configs to use the correct AMPLFI architecture and objects
4. Changes how the kafka producer gets setup because the current approach broke for server="local".
5. Fixes a small bug with snapshotter.full_psd_present. The update_size being passed was in units of seconds, but was compared against units of samples. Additionally, the check against full_psd_present in main was slightly wrong due to the subtraction of update_size. Removing it entirely doesn't cause any harm; it just means that contiguous_update_size will be incremented once more before stopping.

[wbenoit26]

Correction: no arrakis-specific token is needed on the cluster. I've got this running right now, but LLPIC/MDC data isn't available through arrakis yet.

[bhgupta20]

There was a minor issue, but that version no longer exists, so it looks good to me!