[AWS] Fix best disk tier on AWS to use the best IOPS

[romilbhardwaj]

AWS supports upto 16000 IOPS and 1000MB/s throughput on GP3. We should use those specs for high disk tier. Makes a significant difference for reading models/datasets.

Benchmarks:
Before, --disk-tier best on AWS:

(storage-demo, pid=29191) ##### Sequential Read Results #####
(storage-demo, pid=29191) EBS:  498.37 MB/s     7604.55 IOPS
(storage-demo, pid=29191) 
(storage-demo, pid=29191) ##### Sequential Write Results #####
(storage-demo, pid=29191) EBS:  485.97 MB/s     7415.25 IOPS

After, --disk-tier best on AWS:

(storage-demo, pid=30375) ##### Sequential Read Results #####
(storage-demo, pid=30375) EBS:  1395.38 MB/s    21291.75 IOPS
(storage-demo, pid=30375) 
(storage-demo, pid=30375) ##### Sequential Write Results #####
(storage-demo, pid=30375) EBS:  1215.33 MB/s    18544.43 IOPS

[Michaelvll]

The previous IOPS was selected to keep the HIGH disk tier's performance for all clouds similar to each other. Should we instead create a new DiskTier.ULTRA for the max possible IOPS?

[romilbhardwaj]

I see. How about we change DiskTier.BEST to the actual best possible disk tier?

[Michaelvll]

Right now, BEST is the best DiskTier among LOW, MEDIUM, HIGH provided by SkyPilot right now (some clouds may not have HIGH tier, so BEST will map to MEDIUM in that case).

I am fine with changing the semantics of BEST to the actual best disk available on a specific cloud. This is related to #3585 and #3517.

[Michaelvll]

This should be fixed by #3860. Closing for now.