Return invalid request for macOS vGPU

[jarugupj]

Summary

• return HTTP 400 invalid_request when a macOS instance request includes a vGPU
• reject the request before checking GPU capacity, so it is not reported as insufficient resources
• leave other GPU creation errors unchanged

Validation

• Linux and macOS tests pass in CI
• local build, vet, and focused error-response tests pass

Scope

This only changes instance creation. Starting an existing instance is unchanged.

---

Note

Low Risk
Scoped to instance-create validation and error mapping on macOS; Linux vGPU behavior unchanged and existing start paths are untouched.

Overview
Instance create now treats vGPU profile requests on hosts without mediated GPU support as a client error instead of a generic or capacity failure.

Adds devices.Capabilities() (Linux reports vGPU support; macOS does not) and moves ErrVGPUNotSupportedOnMacOS to a shared sentinel in devices/errors.go. Create rejects a non-empty GPU profile when SupportsVGPU is false before ReserveAllocation, so macOS no longer surfaces this as insufficient resources. If CreateMdev still returns the macOS unsupported error, wrapCreateMdevErr wraps it as ErrInvalidRequest, which the API maps to HTTP 400 with code invalid_request and the message invalid request: vGPU (mdev) is not supported on macOS.

Tests cover platform capabilities, early rejection (no reservation), error wrapping, and the create-instance error mapping.

^{Reviewed by Cursor Bugbot for commit 9034ff4. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit c0eac5f. Configure here.}

[sjmiller609]

lib/instances/create.go:100 — could we model unsupported vGPU as a device/host Capability (find existing pattern in codebase) rather than branching on runtime.GOOS here? something like devices.Capabilities().SupportsVGPU, implemented per platform, is the more standard way to handle platform specific capabilities

[jarugupj]

good callout - implemented the existing pattern we rely on in the codebase currently!

[sjmiller609]

perfect, thanks!