Initial Commit: Add vultr support

[Bihan]

No description provided.

[jvstme]

@Bihan, thanks for the PR. I looked through it briefly and left some comments, please feel free to respond to them if you have the time. I may add a few more comments later though

[jvstme]

@Bihan, thanks for the update. I looked through the changes and the resulting catalog (python -m gpuhunt vultr --output vultr.csv) and found a few more edge cases that seem to yield incorrect results. Hope you can address these too

[jvstme]

@Bihan, thanks for the update, there are just a few comments left before this PR is good to merge.

[Bihan]

@jvstme I have modified get_bare_metal_plans function as below:

def get_bare_metal_plans(plan: dict, location: str) -> Optional[RawCatalogItem]:
    gpu_details = BARE_METAL_GPU_DETAILS.get(plan["id"], None)
    return RawCatalogItem(
        instance_name=plan["id"],
        location=location,
        price=plan["hourly_cost"],
        cpu=plan["cpu_threads"],
        memory=plan["ram"] / 1024,
        gpu_count=gpu_details[0] if gpu_details else 0,
        gpu_name=gpu_details[1] if gpu_details else None,
        gpu_memory=gpu_details[2] if gpu_details else None,
        gpu_vendor=get_gpu_vendor(gpu_details[1]) if gpu_details else None,
        spot=False,
        disk_size=plan["disk"],
    )

I think this solves the above issues.

Regarding L40s, it is still not available to verify it experimentally. Currently I have considered vbm-64c-2048gb-8-l40-gpu as L40s.

[jvstme]

@Bihan, thanks, I added a couple more suggestions - please see if you agree with them and we can merge the PR then.

[jvstme]

👍👍