[GCP] Remap series-specific disk types

[JiangJiaWei1103]

Tracking issue

Closes #4705.

Why are the changes needed?

GCP instances have series-specific disk type supports which are summarized as follows:

Machine series	pd-standard	pd-balanced	pd-ssd	pd-extreme
N2	V	V	V	V (>= 64 vCPUs)
N1	V	V	V	—
N1+GPU	V	V	V	—
A2	V	V	V	—
G2	—	V	V	—

To make SkyPilot smart at figuring out the best disk type for each specific machine series, we consider remapping disk types for those unsupported, other than raising an error.

What changes were proposed in this pull request?

• Propagate HIGH disk tier (which maps to pd-ssd) to ULTRA as N2 (with less than 64 vCPUs), N1, A2, and G2 don't support pd-extreme
• Propagate MEDIUM disk tier (which maps to pd-balanced) to LOW as G2 doesn't support pd-standard

Screenshots

• sky launch -t n2-standard-32 --disk-tier ultra --dryrun

• sky launch -t n2-standard-128 --disk-tier ultra --dryrun

• sky launch -t n1-standard-32 --disk-tier ultra --dryrun

• sky launch -t g2-standard-4 --disk-tier ultra --dryrun

• sky launch -t g2-standard-4 --disk-tier low --dryrun

Because we refactor the propagation logic, experiments on A3 series are rerun (see #5351):

• sky launch -t a3-highgpu-8g --disk-tier ultra --dryrun

• sky launch -t a3-highgpu-8g --disk-tier low --dryrun

Tested (run the relevant ones):

• Code formatting: install pre-commit (auto-check on commit) or bash format.sh
• Any manual or new tests for this PR (please specify below)
• All smoke tests: /smoke-test (CI) or pytest tests/test_smoke.py (local)
• Relevant individual tests: /smoke-test -k test_name (CI) or pytest tests/test_smoke.py::test_name (local)
• Backward compatibility: /quicktest-core (CI) or pytest tests/smoke_tests/test_backward_compat.py (local)

[JiangJiaWei1103]

@SeungjinYang PTAL, thanks a lot!

[SeungjinYang]

Nice! I assume you might not be done with the PR given it is in draft state, but this looks good. I was a bit worried the logic might be a sprawl, but the extra logic added ends up being compensated by the logic sprawl removed in check_disk_tier.

[JiangJiaWei1103]

After applying all the suggested changes to the code, we re-ran the experiments and obtained the same results:

• sky launch -t n2-standard-32 --disk-tier ultra --dryrun

• sky launch -t n2-standard-128 --disk-tier ultra --dryrun

• sky launch -t n1-standard-32 --disk-tier ultra --dryrun

• sky launch -t g2-standard-4 --disk-tier ultra --dryrun

• sky launch -t g2-standard-4 --disk-tier low --dryrun

• sky launch -t a3-highgpu-8g --disk-tier ultra --dryrun

• sky launch -t a3-highgpu-8g --disk-tier low --dryrun

Btw, I currently skip a2's tests because there's no available VMs to provision.

[SeungjinYang]

Thanks @JiangJiaWei1103! Based on how the code is written I'm sure a2 type works with ULTRA works if n1 type works, which you tested.